DE112014001413T5 - Quadrature error detection and correction - Google Patents

Abstract

In one example, a system and method for detecting and correcting an error in a quadrature receiver (QR) are disclosed. The QR may include a receiver channel configured to divide a received RF signal into I and Q channels. The receiver channel can be sources of error such. For example, in the order prediming error (PD error), LO mixer error and baseband error (BB error). Test tones may be expired on the receiver channel at a plurality of test frequencies, and quadrature error correction may be provided to detect the error from each source. After receiving an RF signal, the quadrature error correction can apply correction coefficients to correct each source of error in reverse order (BB, LO, PD).

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application claims priority over US Provisional Application 61 / 786,393, filed March 15, 2013, entitled "Device for Countering IQ Imbalance in to RF Receiver" and to US Provisional Application 61 / 786,569 filed March 15, 2013 2013 entitled "Method and Device for Identifying Receivers IQ Imbalance", both of which are incorporated herein by reference in their entirety.

AREA OF REVELATION

This application relates to the field of communications, and more particularly to a system and method for detecting and correcting quadrature errors.

BACKGROUND

The Quadrature Amplitude Modulation (QAM) is a modulation technique that z. B. can be used in quadrature receivers. A received signal may be divided into two carrier waves that are 90 degrees out of phase with each other and thus referred to as quadrature carriers or quadrature components. Using QAM arbitrarily high spectral efficiencies can be achieved with QAM by establishing an appropriate constellation size limited only by the noise level and the linearity of the communication channels.

In an ideal RF quadrature receiver, in-phase (I) and quadrature (Q) branches are perfectly normal to each other and have the same gain over frequency. Any imbalance of this ideal produces unwanted images as signals are partially leaked to their frequency-negated counterparts. This means that a signal x Hz generates an image over a local oscillator (LO), x Hz is perceived as LO, and at the same time a signal x Hz under LO generates an image which is perceived as x Hz via LO. This can have serious effects in a wideband multi-carrier direct conversion receiver, as these images fold directly into the baseband. Unless corrected, these images reduce the signal-to-noise ratio (SNR) and thus the sensitivity of the receiver. The most serious effects occur when amplitudes of the received signals are highly different; the loss of SNR is enhanced when a weak received signal is subject to interference from the image of a strong blocker.

BRIEF DESCRIPTION OF THE DRAWINGS

The present disclosure is best understood by the following detailed description taken in conjunction with the accompanying drawings. It is emphasized that according to standard industry practice, various features are not to scale and are used for purposes of illustration only. In fact, the sizes of the various features can be arbitrarily increased or decreased for clarity of explanation.

1 FIG. 4 is a block diagram of a quadrature receiver according to one or more examples of the present description.

1A FIG. 13 is a series of error representations for error sources in a quadrature receiver according to one or more examples of the present description. FIG.

2 FIG. 4 is a block diagram of quadrature error correction according to one or more examples of the present description.

3 FIG. 4 is a block diagram of a method for detecting quadrature errors in polar form according to one or more examples of the present description.

4 FIG. 10 is a block diagram of a method for detecting quadrature errors in a rectangular shape according to one or more examples of the present description.

5 FIG. 10 is a block diagram of an error observer and error analyzer according to one or more examples of the present description.

6 FIG. 3 is a flowchart of a method of performing frequency domain averaging (FDA) according to one or more examples of the present specification.

DETAILED DESCRIPTION OF THE EMBODIMENTS

OVERVIEW

In one example, a system and method for detecting and correcting an error in a quadrature receiver (QR) are disclosed. The QR may include a receiver channel configured to divide a received RF signal into I and Q channels. The receiver channel can be sources of error such. For example (in sequence), pre-demodulation errors (PD errors), LO mixer errors and baseband errors (BB errors). Test tones may be driven on the receiver channel at a plurality of test frequencies, and quadrature error correction may be provided to detect errors from each source. After receiving an RF signal, the quadrature error correction can apply correction coefficients to correct each source of error in reverse order (BB, LO, PD).

In one embodiment, an integrated circuit for correcting quadrature errors in a received signal, comprising circuitry and logic configured to receive a separate error correction coefficient for each of a plurality of sequential error sources; and apply the error correction coefficients to the received signal in a reverse order of the sequence of error sources.

In another embodiment, a quadrature receiver is disclosed that includes a receiver channel configured to receive a radio frequency (RF) signal; and a quadrature error correction configured to receive a separate error correction coefficient for each of a plurality of sequential error sources; and apply the error correction coefficients to the RF signal in a reverse order of the sequence of error sources.

In another embodiment, a method of detecting quadrature errors in a received signal is disclosed, comprising receiving a separate error correction coefficient for each of a plurality of sequential error sources; and applying the error correction coefficients to the RF signal in a reverse order of the sequence of error sources.

In another embodiment, there is disclosed an integrated circuit for detecting quadrature errors in a received signal comprising a circuit and logic configured to divide the received signal into I and Q channels, the I channel and the Q-channel are separated by a nominal phase angle; process the signal in a plurality of stages, at least some of the stages comprising sources of error characterized by concentrated error models; and calculate an individual error value comprising a gain and a phase for each error source according to the lumped error model for that error source.

In another embodiment, there is disclosed a quadrature receiver including a quadrature receiver configured to receive a radio frequency (RF) signal; and a quadrature error correction configured to divide the received signal into I and Q channels, wherein the I channel and the Q channel are separated by a nominal phase angle; process the signal in a plurality of stages, at least some of the stages comprising sources of error characterized by concentrated error models; and calculate a single error value comprising a gain and a phase for each error source according to the lumped error model for that error source.

A method for detecting quadrature errors in a received signal, comprising dividing the received signal into I and Q channels, wherein the I channel and the Q channel are separated by a nominal phase angle; process the signal in a plurality of stages, at least some of the stages comprising sources of error characterized by concentrated error models; and calculate a single error value comprising a gain and a phase for each error source according to the lumped error model for that error source.

EXEMPLARY EMBODIMENTS OF THE DISCLOSURE

The following disclosure provides many different embodiments or examples for implementing various features of the present disclosure. Specific examples of components and arrangements are described below to simplify the present disclosure. Of course these are just examples and should not be limiting. Further, the present disclosure may repeatedly use reference numerals and / or letters in the various examples. This repetition is for the purpose of simplicity and clarity and in itself does not determine any relationship between the various illustrated embodiments and / or configurations.

Different embodiments may have different advantages and no particular advantage is required of any embodiment.

In an exemplary QE correction microarchitecture (QECM), two related but separate receiver (Rx) channel QE correction (QEC) algorithms can be implemented. The QECM may include a direct mix receiver offering up to 200 MHz baseband bandwidth. An exemplary digital algorithm determines analog sources of quadrature error (QE) in stages by source, and then provides phase and gain correction digitally over frequency, again in stages in reverse inverse order of the source. By counteracting QE in stages opposite to the error alignment, the correction is achieved effectively and accurately. Advantageously, in one embodiment, because of the similarity of the computations required, most of the QE observation hardware and overall QE correction hardware can be shared by the two algorithms; only one QE observation algorithm needs to be active at a time. Both algorithms observe the QE in the frequency domain, map it to a concentrated analog source to still identify the appropriate compensation in the frequency domain, and then apply the correction to the IQ signal in the time domain.

The first algorithm uses an on-chip RF test tone generator to directly observe the QE over the frequency. This algorithm requires the receiver to be set offline to prevent the received signal from interfering with the RF test tone. The second algorithm requires the receiver to be online because it identifies the IQ imbalance based on an unexpected statistical correlation between positive and negative frequencies. An inherent limitation of this second algorithm is that QE monitoring for a particular frequency can not begin until a receive signal is provided at that frequency.

Together, these two calibration algorithms work together to meet the QR calibration requirements. For the most part, they counteract the weaknesses of others. HF tone calibration is typically used during startup, but also supports custom recalibration. It is also used to calibrate a feedback path that is important for channel calibration transmission. In the meantime, the statistically based algorithm is notable for non-invasive calibration tracking. However, to speed up the convergence, it benefits from the notification of the initial full channel calibration requirements, which can be identified by an initial RF tone based calibration. The statistically based calibration can also be used alone in situations in which it can not be tolerated to take the receiver offline. Reasons for this may be related to cost, as an external T / R switch is required to achieve adequate antenna isolation during RF tone based calibration. Reasons for this may be related to the system, as it may not be acceptable to occasionally take the recipient offline. Setting the receiver offline may be tolerable in a TDD setup, but it could be problematic for a FDD setup.

• a. Ton-Nicht-Idealitäten – Tonfrequenz unterscheidet sich von der erwünschten und weist zeitvariierende Phase und Amplitude auf.
• b. Kanalkontaminierung – Unerwünschter Signalinhalt wie z. B. empfangenes HF-Signal ist zusätzlich zum angelegten Testton vorhanden. Idealerweise ist der Testton die einzige Signalquelle.

In a system that contributes to meaningful observation of errors due to their analog or digital construction, the error in IQ imbalance observation is dominated by:

• a. Tone-non-idealities - Tone frequency is different from the desired one and has time-varying phase and amplitude.
• b. Channel Contamination - Undesirable signal content such. B. received RF signal is present in addition to the applied test tone. Ideally, the test tone is the only signal source.

During IQ imbalance monitoring, the received RF signal and other inputs to the receive path should ideally be amplified relative to the applied test tone by using a Switch (inside or outside the device) is disabled, the LNA is overridden, and a minimum input gain setting or other available option is selected.

RF test tones do not have to be pure sine waves with high frequency and phase accuracy.

The purity of the test tone improves the possible rate of convergence, but moderate errors from the intended frequency and phase can be counteracted without limiting the degree of accuracy of the observed IQ imbalance. Also, overtones and undertones are acceptable and actually provide a way to calibrate multiple frequency regions simultaneously.

Frequency errors and phase noise may be erased over time provided they are randomly distributed and have a zero mean error as perceived by the disclosed observation methods of this specification.

It is difficult to completely eliminate short-term audio frequency drifts and phase noise when the frequency accuracy over time is achieved using a feedback loop. The use of a fluctuating feedback loop, which is typically provided by a PLL, shapes the error distribution favorably for quenching by the QEC 170 Not.

The sound analysis performed in the frequency domain evaluates only the subband containing the calibration reference. This provides some immunity to consequences that might otherwise result from the interference. The root of the calibration tone can be used as the calibration reference. Alternatively, an overtone of the calibration tone may be used. The use of an overtone instead of the root is disadvantageous because the calibration reference has less power, but the use of an overtone instead of a root can ease the difficulty associated with generating a high frequency tone.

Multiple subbands can be tested in parallel if multiple tones are created simultaneously. However, tones may not be placed simultaneously in pairs of corresponding positive / negative subbands (i.e., pairs that are offset by LO equidistant but opposite in sign). This would negate the methods of isolating the source of error described in this specification. Many of the overtones and overtones of a sound can not be used in general because it is unlikely that more than one would be within the band.

1 is a block diagram of a quadrature receiver (QR) 100 according to one or more examples of the present description. Throughout the rest of the description will be a specific QR 100 Which is characterized by the integrated circuit available from Analog Devices AD9368 ^®, as an illustrative example uses. It should be noted, however, that AD9368 is provided as a single and non-limiting example, and that many embodiments consistent with the appended claims are possible.

In this example is an antenna 108 configured to receive a radio frequency (RF) signal and the signal to a receiver (Rx) channel 104 via a low-noise amplifier (LNA) 120 provides. The Rx channel 104 includes a plurality of stages, each of which may include a source of error in certain embodiments. In certain configurations, each source of error may be characterized by a concentrated error model, the main focus being on concentrating together sources of error that can be detected and corrected together without specifically referring to which hardware provides the error.

In a predemodulation stage, the RF signal may have a predemodulation error (PD error). 110 receive. The predemodulation error comprises a phase and gain error originating in the error applied to the received RF signal before it is demodulated from the mixer to the baseband or having an identical sequence. In one or more embodiments of this specification, the actual source of error is less important than the symmetry of the corrective action required at the baseband to counteract the error. In the mixing consoles 130 and 132 The RF signal is multiplied by a local oscillator frequency (LO, local oscillator) so that the RF signal can be downgraded to baseband. In one example, the mixer represents 130 an I-channel ready, and the mixer 132 provides a Q-channel, the z. B. is rotated 90 degrees. There is disclosed herein a phase angle of 90 degrees, and this is a conventional value used in many industrial devices, but it is to be understood that other phase angles are possible and beyond that 90 degrees is a nominal or ideal phase angle and that any deviation from an ideal 90 degree phase angle can form part of the QE described herein. The channels I and Q can be represented as complex vectors, each with a real amplitude and an imaginary component. Any error introduced at this stage may be considered a LO error 112 being represented.

Channel I can be used for a transimpedance amplifier (TIA) 142-1 and then for a Sigma Delta ADC (analog-to-digital converter) SD 150-1 to be provided.

Similarly, the channel Q for a transimpedance amplifier (TIA) 142-1 and then for an SD 150-2 to be provided. This level can be a baseband error 114 between Q and I.

1A provides graphical examples for each of a PD error 110 , a LO error 112 and a BB error 114 ready. Depending on where an error originates in the signal chain, this affects frequencies above and below LO with different symmetry. 1A graphically illustrates the even and odd symmetries associated with an amplifier error and a phase error occurring in the stages of PD 110 , LO 112 and BB 114 be introduced.

The QE as described herein relates to the antenna as compared to the original RF signal 108 rather than an absolute error, rather a mismatch between channels I and Q from the relationship expected for an IQ receiver. Thus, in one embodiment, an RF test tone 102 provided with a known amplitude and a known angle, and I _E 182 and Q _E 192 can participate in the QE correction (QE corrector, QEC) 170 to determine a baseline or a calibrated value for QE. It is a switch 160 provided to the application of the HF test tone 102 to the Rx channel 104 to control.

The QEC 170 is described hereinbelow with greater accuracy, but in general the QEC 170 be designed to measure the QE by the source using a test tone 102 to detect and the QE z. B. mathematically correct by applying a correction in a reverse sequence to the error sources. So can the QEC 170 z. B. QE values for the PD error 110 , the LO error 112 and detect and calculate the BB error separately, and it can correct in the order BB error 114 , LO error 112 and PD errors 114 provide. It should be understood that these steps are provided by way of non-limiting example only and that certain embodiments may detect and correct other sources of error, such as those described with reference to FIG. For example, by way of non-limiting example, an error in TIAs 142 or in SDs 150 , Identifying the error by the concentrated source and correcting the error in the reverse order in which it is generated has a significant advantage. As will be described in greater detail hereinbelow, this allows the error to be addressed with a higher degree of accuracy with less computational effort. The obvious complexity of the correction that is required is significantly reduced, if not related error sources, which are measured by their symmetries, as in 1A is counteracted before they interact. Issues of the QEC 170 comprise corrected I and Q signals I _C 180 and Q _C 190 ,

As in 1 can be illustrated, the QE for the quadrature receiver 100 in one example to be concentrated in three sources: PD error 110 (Mismatch in the predemodulation stage), LO error 112 (Mixer LO associated error) and BB error 114 (BB-channel mismatch errors). The LO error 112 arises when the between the mixer 130 and the mixer 132 applied phase angle is not exactly 90 degrees. In this concentrated model, there is the BB error 114 from the residual Rx error after demodulation plus all frequency independent amplitude mismatches before demodulation. In some examples, the frequency independent amplitude mismatch has an indistinguishable effect regardless of the source, at least as far as the correction is concerned, so that it is practical with the BB error 114 too concentrated. The BB error 114 is due to the phase and amplitude mismatch within TIA 142-2 and continuous-time sigma-delta (-SD) 150-2 dominated by mismatched gain expressions within the mixers 130 and 132 , The last concentrated source, the PD error 110 , is the lesser of the three and, in some embodiments, becomes significant only when the bandwidth is large. He can z. By an unbalanced application of LO 140 inside the mixing consoles 130 and 132 , so z. B. Unequal LO duty cycle generated. This tends to produce an asymmetric amplitude mismatch with the frequency of LO 140 is linear.

In the example QR 100 For example, the QE correction is focused solely on eliminating the difference between the I and Q channel responses. This allows I to be considered as a reference, and any mismatches can be focused on Q. Correcting QE then involves identifying of amplitude and phase mismatching of Q with respect to I over frequency and modifying the Q reaction to match that of I identified. Thus, in one or more embodiments, either only one of I _E 182 or Q _E 192 be considered as a meaningful "error" signal requiring correction. In the particular embodiment disclosed herein, only Q _E 192 a meaningful error signal.

Amplitude error like the LO error 112 and the BB error 114 affect positive and negative frequency counterparts in an identical manner if they have arisen after demodulation. Therefore, an amplitude error can be corrected by simply amplifying the amplitude of Q so that it matches that of I in a frequency-dependent manner, without considering whether the affected frequency is actually positive or negative. Herein, an RF frequency is considered negative if it is less than the LO frequency and positive if it is greater than the LO frequency. A frequency-independent amplitude error before demodulation can also be solved in this way, since this definition identically affects both positive and negative frequencies. However, the frequency-independent amplitude error before demodulation requires special attention, as explained below.

Unlike amplitude errors, phase errors may require them to be accurately assigned to their sources for proper compensation. This is due to the fact that phase errors in the BB error 114 shift positive and negative frequency counterparts in an identical direction; while through the LO error 112 introduced phase errors shift positive and negative counterparts in an opposite direction. Correcting the LO error 112 and the BB error 114 at the same time across the full range is in QR 100 achieved by using two different corrective countermeasures. Again the PD error 110 pay special attention, as explained below.

An effective method for correcting a LO error 112 is to add (or subtract) a scalar set from I to Q. If one uses the phase relationship between I and Q, this has a favorable effect on the LO phase error 112 by shifting all frequencies by an equal amount of phase, but as required in one direction depending on the sign of the frequency. An effective way to correct a BB error 114 This is a correction only on the Q channel in the QEC 170 to implement. Other than adding I to Q, the QEC 170 shift the phase of positive and negative frequency counterparts in the same direction, namely an amount that depends on the frequency response of the QEC 170 is determined (the actual sign of an output frequency is irrelevant since this information is lost if Q is evaluated or processed independently of I). The QEC 170 can also correct the amplitude of the BB error 114 may be used and may include a scalar expression for correcting the Q amplification gain resulting from the LO error 112 results. In addition to causing the Q to be sampled at the wrong phase, the LO error can 112 also cause Q to be scanned off its tip. In one example, these two corrective actions (adding a signed set from I to Q and applying QFIR 260 of the 2 ) in the opposite order in which the errors which they counteract were introduced. Until the BB error 114 from the QFIR 160-2 is corrected, the relationship between I and Q is not balanced to what it was, as the LO error 112 introduced his artifacts to Q, so that the correction can no longer be as trivial as adding I to Q. Thus, in one example, the QFIR 260 of the 2 before the scalar addition of I to Q.

Once the LO error 112 and the BB error 114 are corrected, a compensation of the PD error 110 achieved easier. For the same reason, it was cheap, the BB error 114 before the LO error 112 to correct, it is also cheap, both the BB error 114 as well as the LO error 112 before the correction of the PD error 110 to correct. In one embodiment, it was observed that the PD error 110 has a relaxed transfer function. The compensation can thus be achieved by a relatively simple complex filter on the Q channel, whereby positive and negative frequencies can get the slight independent adjustment they need.

In an exemplary system, the QR 100 Be part of a transceiver system, and the test tone 102 can on the QR 100 whenever the system is in transmission mode. Thus, the QE in the QR 100 be updated continuously when the QR 100 not actively receiving an RF signal. In other embodiments, additional arrangements are possible. So z. B. in another example, the QR 100 Part of a dual band assigned receiver without transmit functionality. In this case, one of the two receiver bands may occasionally be switched inactive to the test tone 102 to drive and to measure the QE. In another example, the test tone 102 on a received RF signal are superimposed. In yet another example, the received RF signal itself may be used as a test tone, and the QE may be updated continuously. In one embodiment, the test tone 102 comprise a series of corresponding frequency pairs f, each comprising f + f and -f.

vor.In one example, a concentrated error model for LO (f), wherein the evaluation at frequency f comprises combining the observations at frequencies + f and -f, is in the form of

in front.

vor.In one example, a concentrated error model for BB (f), wherein the evaluation at frequency f comprises combining the observations at frequencies + f and -f, is in the form of

in front.

vor.In one example, a concentrated error model for PD (f), wherein the evaluation at frequency f comprises combining the observations at frequencies + f and -f, is in the form of

in front.

Wherein:
LO _{PHA is identified} by fitting LO _PHA (f) to a polynomial of order n and using the returned value, f is set to zero because the value for LO _PHA . LO _{PHA is} the true LO phase error that does not vary with frequency.

In the above equations, PHA (f) denotes the phase mismatch, which is the deviation from the expected difference between the phase of Q and I observed at the frequency f. For a positive tone, it is expected that the phase of Q lags the phase of I by 90 degrees, while for a negative tone one expects the phase of Q to lead the phase of I by 90 degrees.

Also in the above equations, MAG (f) denotes the amplitude mismatch, which is represented as the ratio of the amplitude of Q to the amplitude of I.

2 is a block diagram of a QEC 170 according to one or more examples of the present description. It is understood that many different architectures of QEC 170 are possible, and the embodiment shown herein is disclosed by way of non-limiting example only. In this example, the QEC includes 170 five primary sub-assemblies: an RF tone generator 220 , a QE observer 230 , a QE analyzer 240 , a QEC controller 250 and an error compensator 280 ,

In one example, the tone generator generates 220 an RF calibration signal just before the mixer 130 is summed to Rx. The orthogonality of the resulting signals on I and Q can then be monitored to determine QEs. Using the corresponding positive / negative frequency pairs, the PD error can 110 and the LO error 112 from the BB error 114 be isolated. Multi-frequency test tones are applied to construct a complete profile of IQ mismatch versus frequency and also to provide sufficient independent equations to identify all unknowns. Analysis of the channel response to a single corresponding frequency pair f identifies the LO error if there is no pre-modulation error and no interference. However, additional pairs are necessary to attest to the presence of the pre-modulation error over the frequency or to cause the interference over the frequency so that it can be counteracted. If the LO error and the predemodulation error are sufficiently identified, unilateral test tones (positive or negative only) may be used to construct the remaining BB mismatch profile.

In one example, an error watcher includes 230 Hardware to transform the IQ signal into the frequency domain. The hardware then performs algorithm-specific accumulations and correlations necessary to obtain the observations that can be used to produce a profile of the phase and gain differences between I and Q over the frequency.

In one example, the QE analyzer processes 240 Observations detected by the QE observation subunit to determine the correction coefficients to be used by the QE correction subunit. The analysis is performed by software that resides in an embedded and shared processor such as a computer. B. an ARM M3 is executed. As used throughout this description, the processor includes any combination of hardware, software, or firmware that provides the programmable logic, including, as a non-limiting example, a microprocessor, a digital signal processor, a field programmable gate array, programmable logic Array, an application specific integrated circuit, or a virtual machine processor. A processor may operate in conjunction with a memory comprising any suitable volatile or non-volatile memory technology, including DDR RAM, SRAM, DRAM, flash, ROM, optical media, virtual memory regions, magnetic or tape storage, or any other suitable technology ,

The analysis is generally not time critical; the hardware within the error observer 230 performs all necessary real-time observations and calculations. For tone-based calibration (TCAL), the analysis starts by algebraically combining observations to delete unwanted expressions that contribute to errors. Extreme outliers can be eliminated to clean up all contaminated sound observations. Low pass filtering, both frequency and observation history (time), can also be performed to improve the observation quality. This may be followed by linear or quadratic interpolation or higher order or curve fitting to estimate gain and phase correction required in subbands lacking tone observation. Subband observations may be missing, either z. Because they were eliminated as outliers or because no sound was applied in that subband during calibration. A final inverse fast Fourier transform (IFFT) may be performed to generate time domain correction coefficients. For statistic-based calibration, the analysis involves identifying the least mean square (LMS) matched polynomial that best fits the positive / negative frequency observations and then using IFFT to generate time domain correction coefficients.

In an exemplary error compensator 280 QE is corrected by first applying the Q channel response to I using a true QFIR 260 balanced on Q, then a scalar set of I of Q in the summation block 214 subtracts and finally balances Q to I using a compensating finite impulse response (CFIR) filter 270 completed on Q. The number of taps required can be determined by conservatively identifying the granularity of the frequency domain adjustments required to perform the IQ mismatch transfer function of the QR 100 track. In addition to forcing the Q response to match I, the error compensator may 280 also add a group delay to I to compensate for the delay added to Q during the correction. delay elements 210 and 212 can be used to provide accurate timing for this group delay.

The QEC control 250 may include user-retrievable registers that may be written and read to initiate, configure, and monitor the calibration. This subunit also has rudimentary calibration sequencing control over all other sub-assemblies of the QEC 170 , In one example, the QEC controller 250 may include a separate processor, or it may be with the QE analyzer 240 share a processor.

In one example, the hardware evaluation of QE is based on the DFT or FFT analysis of the test tones. Test tones can thus be placed at or very close to the midpoints of the subbands they are examining. Non-centered test tones can result in a spectral loss over the multiple FFT subbands. This leakage can amplify the sound observation noise, especially when the sound is at a frequency near LO or near the Nyquist frequency. Near Lo or Nyquist, the leakage from the frequency x to -x is greatest due to the circular nature of the Fourier analysis. Licking x to -x is the most problematic, as it is exactly the subband being evaluated. Non-centered tones may interfere with QE identification because unequal discontinuities on the I and Q channels are detected after the DFT or FFT window. Due to the phase difference of 90 degrees between I and Q, essentially different discontinuities are generated which produce substantially different spectral artifacts. These artifacts can not be contiguous with the original signal and have no particular quadrature relationship.

With regard to the consequences of channel contamination, when unwanted (non-sound) channel content is uncorrelated with itself and the test tone over time, it can travel along a 3 dB trend for each doubling in the number of observations made in the Frequency domain are averaged, be reduced. This is due to the difference between the audio and noise autocorrelation. Perfect tones constructively interfere when periods are summed, yielding 6 dB per doubling of the number of summed periods, while the uncorrelated signal interferes less constructively and amplifies only 3 dB per doubling of the summed samples, at least as a trend.

As an alternative to averaging in the frequency domain, the SNR can also be improved by averaging the multiple data segments (with FFT size) in the time domain. This means that data at each offset within a segment can be averaged with data at the same offset in all other segments to be combined. The result can then be moved to the frequency domain using a single FFT transform, thereby avoiding the considerable computations associated with many individual FFTs. If the frequency of the sound is accurate and the channel noise is completely uncorrelated with the test tone, any doubling in the number of summed segments before FFT analysis increases the SNR along a 3 dB trend. This results from a tone amplitude that improvises by 6 dB per doubling of the summed segments, but the noise amplitude improves only along a trend of 3 dB per doubling. The net improvement in SNR is 3 dB per doubling of the number of summed segments. The presence of a pitch error reduces this SNR enhancement because the signal phase captured by sequential segments drifts at a rate proportional to the frequency error. This reduces the correlation between segments. Although summing segments in the time domain is more efficient in terms of computation, most of the required SNR enhancement in the frequency domain may need to be achieved to improve the degree of immunity to the audio frequency error.

The SNR improves with increased FFT or discrete Fourier transform analysis size (discrete, Fourier transform, DFT). In particular, the SNR in dB improves in proportion to 10 log ₁₀ n _points , where n _{points is} the number of points used in the analysis.

Thermal noise at the antenna sets the minimum bound to the theoretical level of the channel noise. In one example, a constructive goal is that QR 100 is not reduced by more than 2 dB (2 dB noise factor). Starting from -174 dBm / Hz thermal noise at room temperature and assuming a gain of 12 dB from the LNA input to the mixer input, the input of the noise level at the mixer is 130 -174 + 2 + 12 = -160 dBm / Hz or -173 dBVrms / Hz assuming a load of 50 ohms. If 200 MHz of the baseband bandwidth is captured using a 48-point FFT, the total noise is per subband -173 + 10log ₁₀ 200/48 = -106 dBVrms , Since the maximum signal level in mixer input is -17 dBVrms, the maximum signal-to-noise ratio is -17 - (-106) = 89 dB when a full-scale tone is applied. With certain votes in the design, in one exemplary embodiment, within QR 100 a fullscale tone is only possible for a transmission feedback path. For the Rx channel 104 the sound is limited to a full scale of -20 dB, which reduces the signal to noise ratio to 69 dB. The signal-to-noise ratio for the receive path may be dominated by poor antenna isolation rather than thermal noise. Disabling the LNA and opening a T / R circuit should provide an antenna isolation of 50 dB, where the tone-to-tone ratio (sound antenna interference) would be -20 - (-50) = 30 dB if a fully scaled signal from the antenna 108 within a subband of FFT size.

After generating suitable error models according to the TCAL method disclosed herein, the QEC 170 Generate correction coefficients for correcting received RF signals. The correction coefficient may use the concentrated error models. In one example, the correction coefficient may be in an error compensator 280 of the 2 be performed.

The mathematical basis for the generation of the coefficient follows.

The mathematical basis for error observation is given below:

Adoption:
HF is demodulated by mixing with cos (2 * tF _LO ) on I. AND
HF is demodulated to Q by mixing with cos (2 * tF _LO + * 2 + E _LO ) = -sin (2 * tF _LO + E _LO )

Where: F _LO = LO frequency E _LO = LO phase error (total phase of LO for Q is · 2 + E _LO ) t = time If an HF tone of frequency T _{HF is} applied to the input of the mixer, the result after low pass filtering is: I = ½sin (2 × t × (T _HF -F _LO )) Q = ½sin (2 x t x (T _HF - F _LO) - x 2 - E _LO) Substituting T _BB = abs (T _HF - F _LO),
then applies to positive frequencies (T _HF > F _LO ) I = ½sin (2t × T _BB ) Q = ½sin (2t × T _BB - / 2 - E _LO ), while for negative frequencies (T _HF <F _LO ) I = ½sin (-2tT _BB ) = ½sin (2tT _BB +) Q = ½sin (-2tT _BB - 2 - E _LO ) = ½sin (2tT _BB + + / 2 + E _LO ) = ½sin (2tT _BB + 3/2 + E _LO ) If I is multiplied by Q *, the result is:
an amplitude equal to {{i} × mag {Q} AND
a phase equal to pha {I} - pha {Q} (since the complex conjugate of Q was used).

If we divide this product by ² {Q [f]}, then the result will have exactly the amplitude required for baseband amplitude correction, which may be {I} / mag {Q}

It should be noted that the error occurred immediately after the mixer and before the baseband error: For positive frequencies: pha {I × Q *} = pha {I} - pha {Q} = (0) - (- · / 2 - E _LO ) = · / 2 + E _LO For negative frequencies: pha {I × Q *} = pha {I} - pha {Q} = (·) - (3 · / 2 + E _LO ) = - · / 2 - E _LO Each baseband error affects both positive and negative frequencies in the Q channel equally. So the phase of correlation becomes as follows: For positive frequencies: pha {I × Q *} = pha {I} - pha {Q} = · / 2 + E _LO - E _BB For negative frequencies: pha {I × Q *} = pha {I} - pha {Q} = - · / 2 - E _LO - E _BB If instead of I × Q * I * × Q is used for only negative frequencies: For negative frequencies: pha {I * x Q} = pha {Q} - pha {I} = (3 · / 2 + E _LO + E _BB ) - · = · / 2 + E _LO + E _BB If · / 2 is then subtracted from each of the above: For positive frequencies: pha {I} - pha {Q} - · 2 = · / 2 + E _LO - E _BB - · / 2 = E _LO - E _BB For negative frequencies: pha {Q} - pha {I} - · / 2 = · / 2 + E _LO + E _BB - · / 2 = E _LO + E _BB The above expressions are useful because:
Adding their phases can be used to isolate E _LO
Subtracting their phases can be used to isolate E _BB

There are disclosed herein two error separation methods, which are referred to as method 1 and method 2 for simplicity. Method 1 requires fewer calculations than the method below, but it is sensitive to frequency-dependent errors in predemodulation. The mathematical basis for method 1 follows: Assumption: C [f] = M [f] × e ^ j (E _LO - E _BB [f]) for f> 0, and M [f] × e ^ j (E _LO + E _BB [f]) for f <0

Where: C [f] is calculated on the basis of I [f] × Q * [f] / mag ² {Q [f]} for f> 0, OR I * [f] × Q [f] / mag ² {Q [f]} for f <0 derived as above
ELO = LO phase error (when the total phase of LO for Q = · / 2 + E _LO )
E _BB [f] = BB phase error, which is evaluated at the frequency f
M [f] = mag {C [f]} = mag {I [f]} × mag {Q [f]} / mag ² {Q [f]} = mag {I [f]} / mag {Q [ f]}

Assuming M [+ f] = M [-f] and E _BB [+ f] = E _BB [-f]: (C [+ f] + C [-f]) / 2 = (M [+ f] × e ^ j (E _LO - E _BB [+ f]) + M [-f] × e ^ j (E _Lo + E _BB [-f])) / 2 = (M [f] × e ^ j (E _LO - E _BB [f]) + M [f] × e ^ j (E _LO + E _BB [f]) ) / 2 = M [f] × e ^ jE _LO × (e ^ j (-E _BB [f]) + e ^ j (E _BB [f])) / 2 = M [f] × e ^ jE _LO × cos (E _BB [f]) and (C [+ f] + C [-f] *) / 2 = (M [+ f] × e ^ j (E _LO - E _BB [+ f]) + M [-f] × e ^ j (E _LO + E _BB [-f])) / 2 = (M [f] x e ^ j (E _LO - E _BB [f]) + M [f] x e ^ j (-E _LO - E _BB [f ]) / 2 = M [f] × e ^ -jE _BB [f] × (e ^ j (-E _LO ) + e ^ j (E _LO )) / 2 = M [f] × e ^ -jE _BB [f] × cos (E _LO ) for identifying the frequency-independent LO phase error, extracting the phase of: (C [+ f] + C [-f]) / 2 Then the LO phase correction to be applied is: pha {(C [+ f] + C [-f]) / 2} = pha {M [f] × e ^ jE _LO × cos (E _BB [f])} = E _LO

Although the LO phase error identified by each positive & negative frequency pair should be identical, averaging over multiple frequency pairs can be used to help eliminate the noise included in the observation.

To identify the frequency-dependent BB phase error, extract the phase from: (C [+ f] + C [-f] *) / 2 Then the BB phase correction to be applied is: pha {(C [+ f] + C * [- f]) / 2} = pha {M [f] × e ^ jE _BB [f] × cos (E _LO )} = -E _BB [f] In order to identify the frequency-dependent BB amplitude correction that is required, it should be noted that
if {(C [+ f] + C [-f] *)} / 2 creates the basis,
but this gets worse if there is a LO error, because: may {(C [+ f] + C [-f] *)} / 2 = mag {M [f] × e ^ j (-E _BB [f]) × cos (ELO)} = M [f] × cos (E _LO ) Thus, to correct for this amplitude gain induced by E _LO , {(C [+ f] + C [-f] *)} / 2cos (E _LO ) = M [f] × cos (E _LO ) / cos (E _LO ) = M [f] can be used instead.

The method 2 requires more calculations than the method 1, but it allows the observation of frequency dependent predemodulation errors. The mathematical basis for method 1 is as follows: Assumption: C [f] = M [f] × e ^ j (E _LO - E _BB [f]) for f> 0, and M [f] × e ^ j (E _LO + E _BB [f]) for f <0 Where: C [f] is calculated on the basis of I [f] × Q * [f] / mag ² {Q [f]} for f> 0, OR I * [f] × Q [f] / mag ² {Q [f]} for f <0 derived as above
E _LO = LO phase error (when the entire phase of LO for Q = · / 2 + E _LO )
E _BB [f] = BB phase error, which is evaluated at the frequency f.
M [f] = mag {C [f]} = mag {I [f]} × mag {Q [f]} / mag ² {Q [f]} = mag {I [f]} / mag {Q [ f]} Then: E _LO [f] = ½ (pha {C [-f]} + pha {C [+ f]}) = ½ (atan (imag {C [-f]} / real {C [-f]} ) + atan (imag {C [+ f]} / real {C [+ f]})) And: E _BB [f] = ½ (pha {C [-f]} - pha {C [+ f]}) = ½ (atan (imag {C [-f]} / real {C [-f]} ) - atan (imag {C [+ f]} / real {C [+ f]}))

The values of E _LO [f] should be identical at all frequencies of f, unless E _LO [f] contributes to a frequency-dependent pre-modulation phase error (ie, unlike the name, E _LO [f] is not just the LO error alone associated). As a good approximation, the mean of E _LO [f] may be taken as a mixer LO error, and the variation of E _LO [f] over the frequency of this means may be taken as the frequency-dependent pre-modulation phase error.

Similarly, M [+ f] should be equal to M [-f] unless M [f] contributes to the frequency-dependent predemodulation error. As a good approximation, the average of M [+ f] and M [-f] may be taken for each frequency f as the required BB amplitude error correction. In the meantime, the variation of M [+ f] and M [-f] from the mean at each frequency f may be taken as the amount of the frequency-dependent predemodulation amplitude error.

Regarding the compensation of the BB error 114 A FIR can be applied to Q, as described in the section in the mathematical derivation of paragraph Fehler! Reference source not found. is described. In particular, assuming that the error separation method 2 is used, this filter must apply the amplitude correction: M [f] = ½ (mag {I [f]} / mag {Q [f]} + {I [-f]} / mag {Q [-f]})

The phase correction includes: -E _BB [f] = -½ (atan (imag {C [-f]} / real {C [-f]}) - atan (imag {C [+ f]} / real {C [+ f]} ))

It should be noted that E _{BB is} reversed to change from an error term to a correction term.

It should also be noted that only the mean amplitude correction identified for a particular corresponding positive / negative pair is from the QFIR 260 is served. It is expected that the means will be the required dominant amplitude correction and may be associated with baseband IQ channel mismatching. The smaller remaining amplitude correction that is still needed and as one Asymmetric positive / negative frequency mismatching is mapped from this average to the CFIR 270 relegated. Unlike the QFIR 260 , points the CFIR 270 Both I and Q filter taps have the ability to independently adjust a positive and negative frequency response. The mapping of a corresponding positive / negative frequency amplitude mismatch as asymmetric (but with opposite sign) participation is based on simulation of the mixer and the silicon evaluation of QR 100 justified: a non-perfect duty cycle LO has been verified to contribute to the linear asymmetric amplitude mismatch across the frequency.

zum BB-Korrektur-FIR kann somit verwendet werden, um den LO-Fehler-induzierten Amplitudenverlust zu kompensieren.Although not with the BB error 114 For the BB correction FIR, it is convenient to include a scalar modifier that compensates for the amplitude loss of the Q channel that is associated with the LO error 112 is associated. In accordance with the mixer error model, the mixer error causes the out-of-peak Q channel to be sampled by an amount proportional to cos (E _LO ), where an amount equal to -sin (E _LO ) of I is sampled instead of the missing contribution from Q. Adding a gain term from Q

to the BB correction FIR can thus be used to compensate for the LO error induced amplitude loss.

Regarding the compensation of the LO error 112 For example, in the absence of the other predemodulation phase error, the LO error may be identified using any corresponding frequency pair f. In particular, the following equation would identify the identical LO error regardless of the evaluated corresponding positive / negative frequency pair: E _LO [f] = ½ (atan (imag {C [-f]} / real {C [-f]}) + atan (imag {C [+ f]} / real {C [+ f]}))

If instead this equation calculates different phases for different frequency pairs, these observations can be used to configure the frequency dependent phase correction. In one example, the correction is done using the CFIR 270 implemented. First, however, must be the inherent frequency-independent LO error 112 be separated from the observations about the frequency. This may require a trusted error model to guide the error mappings.

In the absence of a correct error model, the mean derived LO error could be 112 , calculated over the frequency using the above equation, can be used to approximate LO. The phase observed using the lowest frequency pair can provide a better approximation of LO with the assumption of the inherent reduced impact of frequency dependent sources. However, a best approximation can be determined by calculating a DC section. This assumes that the cause of the variation in phase observation is symmetric about the positive / negative separation of DC and that a low order function could be used to adjust these phase observations.

-Skalar umfasst, der vor der Addition von I zu Q angelegt wird (siehe 3), muss auch der LO-Phasenkorrektur-Skalar um diesen Faktor

verstärkt werden. Dies ergibt insgesamt einen LO-Fehler-Kompensationsskalar

welcher derselbe wie tan(E_LO) ist.A convenient way to fix the LO error 112 is to add a scalar set from I to Q. This addition shifts the positive and negative frequencies in the Q channel by equal amounts but in opposite directions. Corresponding to the mixer error model, this scalar is equal to sin (E _LO ). Since the baseband correction FIR but a

-Skalar, which is applied before the addition of I to Q (see 3 ), the LO phase correction scalar must also be around this factor

be strengthened. This results in an overall LO error compensation scalar

which is the same as tan (E _LO ).

An alternative approach to the above digital solution is to directly drop the LO phase in the mixer 130 to correct. If the LO applied to the I and Q channels has a configurable delay, then the initial coarse calibration coupled with a last search for the best quantized step may be used to remove the LO error at HF. This has the advantage that all the upper and lower notes given in HF are treated before changing to baseband, and not just the primary picture being observed at the baseband, without other upper and lower notes Undertones that form aliases compensate. This is especially important when the transmit channel is looped through a calibrated receive channel to facilitate transmission of the QE calibration. Transmitted harmonics and overtones can inherently be amplified minimally in this feedback path.

Regarding the correction of the PD error 110 will be residual QE that can not be counteracted with the above baseband and LO correction procedures by the CFIR 270 served. It is assumed that the remaining amplitude and phase errors over DC are asymmetric for this correction (this is necessary for the phase correction, otherwise other unknowns can not be solved given the number of equations available). To reduce the implied filter order, observations may be adjustments to a low order polynomial such as a first or second order polynomial. Matching to a polynomial reduces the "noise" of the observation error, which can be significant, given the correction applied by the CFIR near the observation boundary. Data points applied to the LMS match are generated by determining the frequency-specific observation deviation from the values chosen for baseband amplitude and LO phase correction, ie deviations from QFIR and the already selected LO compensation.

3 FIG. 10 is a block diagram of error detection and correction according to one or more of the examples of the present description. FIG. The implementation is structured so that hardware can be reused for both sound-based and blind-tracking algorithms.

The calculations of 3 can be sequenced by all released test tones. In one example, at least a single corresponding frequency pair f is enabled to calibrate the LO error, and at least one tone, either positive or negative, must be applied to each subband requiring calibration. The following is done for the released tones.

In the block 320 I and Q can be phase aligned in the time domain. It should be noted that this step is not necessary in all embodiments.

In the block 320 I can be rotated to within an a +/- tolerance of a selected angle such. To be zero degrees, and Q can be rotated by exactly the same amount as I was rotated. This step improves the immunity of the QEC 170 versus calibration tone phase noise, although this is not required in all embodiments.

• a. Für I und Q getrennt Initialisieren eines n-Eingangspuffers mit den ersten n Datenproben. In einem Beispiel n = 48.
• b. Akkumulieren der Puffereingaben mit den nächsten n Datenproben in einer Ringverteilungsweise.
• c. Wiederholen des Schritts (b) bis die Zeitdomänenmittelungsperiode ausläuft. Wenn n = 48, können die Resultate die akkumulierten Proben 1, 49, 97...sein, die akkumulierten Proben 2, 50, 98 und so weiter. Während der letzten Mittelung sequenzielles Ausgeben von n Akkumulationen, um den FFT-Puffer für den Block 324 zu initialisieren. Es muss Spielraum für die maximale Zeitdomänenmittelung, die gestützt wird, bereitgestellt sein. Die Ausgabe kann zurück in die Eingabedatenbreite normalisiert werden (z. B. z. B. durch TDA-Zählung). Alternativ dazu könnte die Ausgabe eine automatische Verstärkungssteuerung aufweisen. Diese Verstärkungssteuerung würde die größten Werte, die für I und Q akkumuliert wurden, überwachen und danach alle akkumulierten Proben von I und Q um eine identische Menge verschieben oder verstärken, so dass alle oder die meisten Bits der Dateneingabebreite mit numerisch signifikanten Bits besetzt wären. Dies bedeutet, dass die Ausgabe bis zu einem Grad verstärkt wird, in welchem zumindest eine akkumulierte Probe von I und Q sich nahe der Sättigung der größten Amplitude befindet, die durch die Breite der Ausgabedaten ausgedrückt werden kann. Diese Technik kann eingesetzt werden, um die Konsequenzen des digitalen Quantisierungsrauschens zu reduzieren, da dieses die Amplitude des erwünschten Signals erhöht. Dies ist besonders wichtig, wenn der Kalibrierungston in Bezug auf eine vollmaßstäbliche Eingabe verstärkt wird.

In the block 322 For example, data segments are overlapped, such as a periodogram, to implement time domain averaging (TDA). This starts the process of improving the SNR for more accurate measurements. The IRR achieved is proportional to the SNR obtained, and the SNR can be determined either by time domain average or frequency domain average measurements as in the Metablock 340 be improved. In some embodiments, the frequency domain average is superior since it can reduce the inaccuracy of the calibration tone frequency and the consequences associated with phase noise. The TDA for this step may include the following:

• a. For I and Q, initialize an n-input buffer separately with the first n data samples. In an example n = 48.
• b. Accumulate the buffer inputs with the next n data samples in a ring distribution fashion.
• c. Repeating step (b) until the time domain average period expires. If n = 48, the results may be the accumulated samples 1, 49, 97 ..., the accumulated samples 2, 50, 98 and so on. During the last averaging, sequentially outputting n accumulations to the FFT buffer for the block 324 to initialize. There must be scope for maximum time domain averaging supported. The output can be normalized back to the input data width (eg, by, for example TDA-counting). Alternatively, the output could have automatic gain control. This gain control would monitor the largest values accumulated for I and Q and thereafter shift or boost all of the accumulated samples of I and Q by an identical amount such that all or most bits of the data input width would be occupied with numerically significant bits. This means that the output is amplified to a degree in which at least one accumulated sample of I and Q is close to the saturation of the largest amplitude, which can be expressed by the width of the output data. This technique can be used to reduce the consequences of digital quantization noise, as this increases the amplitude of the desired signal. This is especially important if the calibration tone is boosted with respect to a full-scale input.

To improve the accuracy of the IQ imbalance observations, frequency domain and time domain averaging can be used.

In the block 324 is a complex DFT or FFT on combined time domain averaged I and Q data from the block 322 executed. The time domain input to the FFT is I + j _Q. The output of the block 324 is a representation of the frequency domain of a sound observation.

m block 330 I and Q are separated from a complex DFT or FFT result. If an FFT is used, only the subband associated with the applied tone is examined: I _REAL [f] = (Y _REAL [f] + Y _REAL [-f]) / 2 I _IMAG [f] = (Y _IMAG [f] - Y _IMAG [-f]) / 2 Q _IMAG [f] = (-Y _REAL [f] + Y _REAL [-f]) / 2 Q _REAL [f] = (Y _IMAG [f] + Y _IMAG [-f]) / 2

In the block 332 For example, I can be rotated to be within a +/- tolerance of a selected angle, such as zero degrees, and Q can be rotated by exactly the amount that I was rotated. This step improves the immunity of the QEC 170 to calibration tone phase noise, but it is not necessary in all embodiments.

In the metablock 340 For example, frequency-domain averaging (FDA) is performed. In the block 342 I × Q * is calculated if the tone is positive (T _RF > F _LO , where T _{RF is} the frequency of the tone and F _{LO is} the LO frequency), or I * × Q is calculated if the tone is negative is (T _RF <F _LO ) where T _{RF is} the RF tone and F _{LO is} the LO frequency and where a tone is positive if it is greater than F _LO and negative if it is less than F _LO is. Whether positive or negative, 90 ° are subtracted from the result. Due to the fact that for a perfect IQ receiver Q for positive frequencies I lags 90 degrees and Q leads for negative frequencies 90 degrees, these calculations produce a complex number with an amplitude equal to the amplitude of I times Q and a phase equal to that required by Q phase matching.

For a positive tone: IQ _IMAG [f] = -I _REAL [f] × Q _REAL [f] -I _IMAG [f] × Q _IMAG [f] IQ _REAL [f] = I _IMAG [f] × Q _REAL [f] - I _REAL [f] × Q _IMAG [f]

For a negative tone: IQ _IMAG [f] = I _REAL [f] × Q _REAL [f] + I _IMAG [f] × Q _IMAG [f] IQ _REAL [f] = -I _IMAG [f] × Q _REAL [f] + I _REAL [f] × Q _IMAG [f]

In the block 344 An amplitude of Q ^{2 is} calculated. Q ² _MAG [f] = Q ² _REAL [f] + Q ² _IMAG [f]

IQ _IMAG [f], IQ _REAL [f] and Q ² _MAG [f] are accumulated in three separate buffers until the frequency domain averaging (FDA) is completed.

As a last step, which is executed after completion of the accumulation, is further in the block 340 the block 342 through the block 344 divided.

Thus: C [f] = FDA {IQ [f]} / FDA {Q ² _MAG [f]}

und eine Phase von f E_LO – E_BB[f].In this example, C [f] is an amplitude compensated IQ correlation. The result C [f] has an amplitude of 0 for positive frequencies

and a phase of f E _LO - E _BB [f].

und die Phase von E_LO + E_BB[f].For negative frequencies, the amplitude is

and the phase of E _LO + E _BB [f].

In one example, the duration of the FDA is the block 340 such, so between this and the TDA of the block 322 the SNR is sufficiently improved to support a target IRR. Both the TDA and the FDA may be programmable for flexibility. Data path travel may be provided for a maximum supported FDA. The output can be normalized back to the width of the input data (divided by the FDA count).

In one example, the metablock becomes 310 , which includes the observation of IQ imbalance, used repeatedly for each test tone. Once the block 310 has processed each test tone, sets the mapping of the IQ imbalance to optimal correction coefficients in the metablock 350 continued. In another embodiment, the metablock 310 partially or completely doubled to support the processing of the multiple tones simultaneously.

In the block 350 Q and I are converted from the rectangular shape to the polar shape.

The block 354 receives a C [f] phase and amplitude and compares observations at + f and -f to map errors into a lumped error model appropriate for the source of the error. This mapping is performed over the full frequency range to provide an array of phase and amplitude error values for each of the PD errors 110 , LO error 112 and BB error 114 to create. Conversion of error terms to correction terms is also done in the block 350 , This involves reversing the sign of the phases and reversing the value of the amplitude errors.

In the block 360 can outliers z. B. can be discarded because they z. B. represent disturbed measurements. Once outliers have been discarded, the remaining values may be matched to an appropriate polynomial of the nth order. These steps are described in more detail below.

In the block 364 , the PD correction coefficients for the error 110 converted from a polar representation style to a rectangular representation style.

In the block 362 will be the BB correction coefficients for the error 114 converted from the polar representation type to the rectangular representation style. The block 362 also receives LO _MAG , which helps in scaling the BB correction coefficients for the error 114 is used to compensate for the amplitude list when the LO error 112 from the adder 214 Will get removed.

In the block 366 the LO error is converted to a scalar correction coefficient by LFIR = tan (LO _PHA ). The LO scalar can be provided to an LFIR. The LO phase error can be z. B. corrected by using a single tap addition from I to Q in the digital domain, as shown, or by adjusting the delay of LO provided to either the I or Q path within the analog mixer an adjustment feedback loop, assuming non-calibrated analogue adaptation steps), or by using a combination of both of these digital and analog methods. Correcting in the analog domain can provide additional benefits, such as: B. the reduction of predemodulation errors. Correcting in the digital domain thereafter, however, still makes sense, since the correction of the remaining error can be made with greater accuracy and ease in the digital domain.

The block 370 performs an IFFT on the right-angle BB correction coefficient for the error 114 by. The output of the block 370 is a time domain representation of the FIR correction filter coefficients for the BB error 114 , Only time domain coefficients are generated due to the symmetric treatment of the positive and negative frequencies within the frequency domain coefficients applied to the IFFT input.

The block 372 performs an IFFT on the right-angle PD correction coefficient for the error 110 by. Both real and imaginary time domain coefficients are generated due to the independent treatment of the positive and negative frequencies within the frequency domain coefficients applied to the IFFT input.

The metablock 380 is a correction block that is the error compensator 280 of the 2 corresponds, with the correction coefficients in the metablock 350 can be calculated to change the response of Q. The metablock 380 receives real correction coefficients for the error 114 , and it receives the scalar correction coefficient for the LO error 112 , The metablock 380 receives both real and imaginary correction coefficients for the PD error 110 ,

In the metablock 380 compensates the QFIR 260 the BB error 114 by adjusting Q only. The CFIR 270 compensates for the PD error 110 , The scalar LO error 112 is compensated by a section of I at the intersection 386 with the issue of the QFIR 260 is summed up.

In terms of the metablock 310 the certainty increases that a single averaged IQ observation is within some accuracy of true IQ imbalance, as the number of averaged observations increases.

In terms of block 360 If multiple subband amplitude observations are obtained prior to the application of a test tone, and if these observations have different values of time and / or frequency domain averaging, then these observations can then be used to determine the SNR (Signal to Noise Ratio) associated with these averaging values and all the values of the averaging in between. The calculation of the SNR also requires knowledge of the sound amplitude. A single measurement of the sound amplitude would inherently include a noise contribution sufficient to degrade the SNR determination. However, the noise can be amplified sufficiently that it is non-dominant if the tone and noise amplitudes used in an SNR calculation are respectively determined by averaging the same number of observations ( s + n / n approaches s / n if n is small with respect to s).

The presence of a continuous wave under a non-correlated signal can interfere with the trend of 3 dB SNR enhancement expected with each doubling in the number of averaged observations. In particular, as soon as the amplitude of the uncorrelated signal contribution is amplified by averaging the amplitude of a continuous wave, there is no further improvement of the SNR. Thus, based on the lack of a 3 dB SNR improvement detected per doubling of the number of averaged observations, the amplitude of a dominant continuous wave can be approximated. This also identifies the favorable limit to the means of observation for a particular subband and the number of observations it pays to accumulate. A basic statistical understanding requires numerous observations at a particular value of applied averaging that is to be captured, but before a sufficient data set is collected to support a given level of security, received signal conditions would likely render that estimate less relevant. Using only a single estimate to roughly approximate SNR may thus be an adequate compromise, especially if a particular backup band is applied to reduce the consequences of inaccuracy.

The potential presence of continuous waves makes the estimation of SNR pessimistic at larger averaging values than the one observed, and the estimate of SNR at smaller averaging values than that observed. The estimates are most accurate if they lie between two observations and do not extend outside the observations.

Observations with different degrees of averaging are efficiently obtained by obtaining a series of 2u intermediate observations, each representing the average of 2v observations, where u and v are positive integers and u is greater than v. These overall observations of 2u x 2v can then be summarized by averaging the doubled portions of these intermediate observations, always starting from the first, to provide information associated with the u values of the doublings.

To best support the prediction accuracy, noise observations can be made just prior to applying the test tones. This improves the likelihood that the signal conditions remain similar and that the prediction remains relevant.

A useful application of a rough estimation of SNR (ie, an estimate obtained after a single averaging period rather than the stronger statistical understanding that results from many such observations) is the identification of a sub-band with a high probability of noise by averaging is not amplified (ie there is noise that has a higher correlation with itself than assumed before).

Subbands that have been observed to have a poor SNR potential can be bypassed for the sound tests, and adjacent subbands with better SNR can be used instead. Moving the tested subbands will only lead to inaccuracy if this results in a region of fast transfer function transition being underrepresented.

In an exemplary embodiment, to accelerate the calibration during repetition across the block 310 only essential sounds created. In particular, tones with greater frequency separation may be spaced apart in regions where a rapid change in IQ imbalance is not expected, or where the accuracy of the IQ imbalance observations is less important (this information could be provided by the user, or by observation determined).

In some embodiments, tones are required only in frequency regions where the user requires correction. The exception to this depends on the determination of the LO error 112 together. Regardless of the actual user correction requirements in the vicinity, the LO error will 112 is best observed near DC (using at least a pair of corresponding positive / negative tones f). This is due to the fact that near DC the receiver had no opportunity to produce a significant IQ imbalance, which, if mispredicted, the accuracy of the LO error observation 112 could disturb.

In one embodiment, three corresponding tone pairs f near DC are obtained, around the LO error 112 appreciate. This allows two out of three pairs to be averaged to estimate the LO error 112 in the case that a pair implies a significant outlier value. This method can be extended to use additional tone pairs beyond the only three, as required.

In some cases, the application of the tones in pairs to individual tones may be shortened as soon as sufficient observations have been made to adequately address both the LO error 112 as well as to identify the transfer functions associated with both odd symmetric amplitude and odd symmetrical phase errors of the PD error 110 are associated.

In terms of blocks 322 and 340 For example, TDA and FDA may be used to improve signal-to-noise ratio along a 3 dB trend per doubling of averaged observations or an equivalent doubling of the number of data samples included in each observation when the channel noise is with both itself and uncorrelated to the test tone. Herein, an observation refers to the study of a single FFT result. A proportionally lower trend is obtained when the channel noise is not completely uncorrelated with itself and the test tone.

The TDA is more computationally efficient than the FDA (fewer FFTs are required) and can therefore provide greater energy and cost savings.

A technique may be used to improve the effectiveness of the TDA when there is thus calibration tone phase noise. If data segments are blind overlapped and summed like a periodogram, calibration tone phase noise can accumulate over time and result in a reduced or even eliminated benefit of averaging. This is because uncorrelated noise increases in magnitude along a 3 dB trend per doubled observation as the tone strength increases with a dependence on the phase alignment of the overlapped and time domain averaged tone periods. In particular, when two tone periods are overlapped / summed, the tone strength increases by ₁₀ log ₁₀ (2 + 2 cos A) dB, where A is the phase misalignment between the two tone periods. If z. For example, if tones are captured from two phase alignment groups, with one group out of phase with the other by 90 degrees, the tone-to-noise power would not experience an improvement from the observation doubling. Perfect tone alignment results in a 3 dB SNR with each doubling of the observations, but phase noise combined with depth averaging can reduce or even eliminate the overall benefit of averaging unless phase noise is counteracted.

An inexpensive means of controlling the audio phase alignment is to include a circuit that requires a minimum signal edge to be detected before a next audio period is captured and overlapped / summed. Assuming that the calibration tone is a sine wave, the flank is greatest at zero crossings, so a minimum flank detection check can be used to ensure that all tone periods that are overlapped / summed are within a selected +/- tolerance of zero. Degree phase lie. Including sound periods with edges with leading edges that are either sufficiently positive or negative is worthwhile as this can reduce the waiting time on edges that will eventually arrive and be sufficient. However, if both edge polarities are used, the sign of the data collected after a negative edge is reversed before summing. Otherwise, a sound erasure rather than an audio gain is achieved. Monitoring the maximum tone slope or tone strength can be used to calibrate the edge boundaries.

The FDA provides an overall rejection of Tone-Non-Idealities (Tone Phase Error can not accumulate and reduce Tone Autocorrelation), therefore it supports averaging lower than Time Domain Averaging.

With respect to the removal of outliers and the polynomial fit of the block 360 may, after observations associated with a single sweep of selected tones (reduced from full speed calibration setting), be obtained using a selected average degree (to reduce the standard error error to an acceptable level), the accuracy of the identified IQ Imbalances are still improving more. Based on the expectation that the true IQ imbalance transfer function is relaxed, strong discontinuities should be eliminated.

Significant outliers (those outside expected IQ imbalance extremes for QR 100 ) can be assigned a strong observation interference and can thus be discarded. So the sound observation z. B. contaminated with correlated noise.

Moderate outliers (those that fail to reach the threshold of significant outliers) may be due to lower levels of observation interference, but may also be associated with device drift or observation variation within statistical norms. Lacking further knowledge of their origin, moderate outliers can be counteracted by combining with previous values per subband using a low pass filter. This limits the immediate response without eliminating the ability to track the drift of the device as a whole.

• a. Ein Bewegungsmittel erster Ordnung (N – 1 vorherige Werte werden mit den aktuellen summiert, wonach die Summe durch N geteilt wird)
• b. Einen verlustbehafteten Integrator (Ausgabe gleich α mal neuer Beobachtung, plus 1 – α mal vorheriger Ausgabe).

Options for low-pass filtering include, by way of non-limiting examples:

• a. A first-order moving average (N - 1 previous values are summed with the current one, after which the sum is divided by N)
• b. A lossy integrator (output equal to α times new observation, plus 1 - α times previous output).

N or α can be chosen to compensate for the ability to track the drift of the device with the desired outlier gain.

• a. Anpassen von Beobachtungen an ein Polynom, wobei der LMS-Fehler (Fehler der kleinsten mittleren Quadrate) minimiert wird. Die Ordnung der Polynome sollte eingeengt werden, um die Komplexität der Übertragungsfunktion zu begrenzen.
• b. Schätzen nicht-beobachteter Subband-Werte auf der Grundlage einer linearen Interpolation zwischen Paaren von Subbändern, die beobachtet wurden, und Anlegen von speziellen Ausdehnungen zu den Nyquist- und DC-Extremen. Insbesondere:
• i. Ausdehnen der beobachteten Phase & Amplitude mit der höchsten Frequenz zu Nyquist. In einigen Fällen begrenzt dies ein wenig eine nicht notwendige Komplexität der geschätzten Übertragungsfunktion.
• ii. Wiederholen der beobachteten Amplituden der niedrigsten Frequenz zu DC. Dies basiert auf einem angenommenen feststehenden I/Q-Verstärkungsfehler bei den niedrigsten Frequenzen.
• iii. Lineares Interpolieren der beobachteten Phase mit der niedrigsten Frequenzhinunter zu null bei DC. Dies basiert auf einer angenommenen linearen Phasenreaktion des Empfängers, zumindest bei einer niedrigen Frequenz.
• iv. Lineares Interpolieren von Phase und Verstärkung zwischen Paaren von Subbändern, die beobachtet wurden.

Once significant outliers have been eliminated, and moderate outliers counteracted, missing subband observations (either due to the removal of outliers or lack of applied test tone) can be approximated to complete an estimate of the complete IQ imbalance transfer function. Values that are able to close these gaps can be determined, either before or after error sources have been separated into concentrated contributions. A variety of curve fitting methods can be used to determine these values, including by way of non-limiting examples:

• a. Fitting observations to a polynomial minimizing the LMS (least mean square) error. The order of the polynomials should be narrowed to limit the complexity of the transfer function.
• b. Estimate unobserved subband values based on a linear interpolation between pairs of subbands that were observed and applying specific extents to the Nyquist and DC extremes. Especially:
• i. Extending the observed phase & amplitude with the highest frequency to Nyquist. In some cases this limits somewhat the unnecessary complexity of the estimated transfer function.
• ii. Repeat the observed amplitudes of the lowest frequency to DC. This is based on an assumed fixed I / Q gain error at the lowest frequencies.
• iii. Linearly interpolate the observed phase with the lowest frequency down to zero at DC. This is based on an assumed linear phase response of the receiver, at least at a low frequency.
• iv. Linear interpolation of phase and gain between pairs of subbands that were observed.

4 FIG. 10 is a block diagram of error detection and correction according to one or more examples of the present description. While the embodiment of the 3 the correction coefficients in polar representation in the Metablock 350 mappt, maps the embodiment of the 4 the correction coefficients in a rectangular representation in the metablock 450 ,

In an embodiment, the block comprises 410 the observation of IQ imbalance and is essentially identical to the block 310 of the 3 and it will look for additional details regarding the block 410 directed. Likewise correspond the blocks 420 . 422 . 424 . 430 . 432 . 440 . 442 and 444 each of the blocks 320 . 322 . 324 . 330 . 340 . 342 and 344 of the 3 , Like the block 310 can also be the block 410 be performed for each test tone.

The block 454 includes a comparison and a mapping, and in one embodiment is substantially identical to the block 354 of the 3 , except that the block 454 performs the comparison and the mapping on rectangular coordinates. It gets on the block 354 of the 3 for additional details regarding the block 454 directed.

The block 460 includes distance and polynomial fitting, and in one embodiment is substantially identical to the block 360 of the 3 , except that the block 460 Distance and polynomial fitting to rectangular coordinates. It gets on the block 360 of the 3 for additional details regarding the block 460 directed.

The block 462 scales the correction coefficients for the BB error 114 around LO _MAG . The block 470 includes an IFFT, and in one embodiment is substantially identical to the block 370 of the 3 , except that the block 470 performs the comparison and the mapping on rectangular coordinates. It gets on the block 370 of the 3 for additional details regarding the block 470 directed.

The block 472 includes an IFFT, and in one embodiment is substantially identical to the block 372 of the 3 , except that the block 472 the comparison and the mapping on orthogonal coordinates. It gets on the block 372 of the 3 for additional details regarding the block 472 directed.

The block 466 includes tan (LO _PHA ), and in one embodiment is substantially identical to the block 366 of the 3 , except that the block 466 performs the comparison and the mapping on rectangular coordinates. It gets on the block 366 of the 3 for additional details regarding the block 466 directed.

The block 480 includes the QE correction, and in one embodiment is substantially identical to the block 380 of the 3 and it will insist on additional details regarding the block 410 directed.

5 is a block diagram of an error observer 230 and a fault analyzer 240 according to one or more examples of the present description. It should be noted that 5 is only one example implementation, which is the same as a quadrature error correction microarchitecture ADI ^® part number AD9368 essentially. This example is not meant to be limiting.

The error observer 230 includes a memory 510 , two FFT blocks 560 , a time domain processor 550 and a frequency domain processor 540 ,

In one example, the time domain processor buffers 550 the segments of the Rx-I and Q data and arranges these, which are then a pair of FFT modules FFT X. 560-2 and FFT Y 560-1 are processed. The FFT modules can be staggered and receive alternating segments. This may be necessary if z. For example, every FFT module 560 is able to process segments at only half the speed at which they are generated. In one example, if the segments are arranged, a custom window is created if the signal * I was detected. Alternatively, the segments may be time-domain averaged before being output, as described by I.

In one example, regardless of the source of the IQ samples, the processing of the samples by the time domain processor 550 adjusted to the speed at which the samples are received. All processing for the time domain for QE analysis is again based on the output rate of the time domain processor 550 customized. This allows IQ samples to QEC 170 at an arbitrary speed and at an interval up to the clock rate of the QEC 170 itself can be supplied.

Segments can be in a virtually addressed memory 510 which is common among numerous modules. The use of a virtually addressed memory 510 allows the immediate transfer of results from one module to another. Individual modules do not require attention to the required address translation.

The error analyzer 240 can optionally be a processor 520 which may be any type of processor described herein. The error analyzer 240 can also have a tone error analyzer 530 include herein as being indirect with the error correction 170 is disclosed connected. In one example, the tone error analyzer 530 with the error correction 170 by a frequency domain processor 540 get connected.

The memory 510 may include a group of k-blocks that are in j-banks 512 are grouped. It should be noted that this configuration, and the entire configuration of the 5 , are disclosed by way of non-limiting example only and that many other hardware configurations are possible.

6 is a flowchart of a method 600 for performing an FDA according to one or more examples of the present specification.

The FDA and the TDA are in blocks 322 respectively. 340 of the 3 shown. A balance between the use of time and frequency domain averaging is appropriate to achieve greatest accuracy with the greatest efficiency. The appropriate ratio depends on the purity of an RF test tone (ie, lack of frequency error and phase noise, which may mimic a frequency error).

In one example, the metablock identifies 340 the required IQ correction, in part by calculating IQ * / QQ * over the frequency. This results in a complex number having a phase equal to the difference between the phase of I and Q and an amplitude equal to the amplitude of Q. From this it is easy to extract the phase and amplitude correction over the frequency. The accuracy of the numerator and the denominator of this calculation, however, does not draw the same benefit from the averaging of several observations; Observation noise is not eliminated equally by averaging. For the counter, uncorrelated noise in I and Q, which has a zero mean, still has a zero mean after I has been multiplied by Q *. For this reason, the averaging can be used to reduce the strength of this noise. However, for the denominator, since Q is multiplied by Q *, the noise is inherent during multiplication, thereby clearing the zero mean. The averaging thus has only limited utility for reducing this particular noise source. The procedure 600 but can be used to eliminate the noise in Q before QQ'-multiplication.

Do you join in? 6 In an example, if in the block 602 two complex accumulators are cleared, the accumulations of the modified I and Q expressions are stored.

In the block 610 a frequency f is selected from an array of n-elements of the test frequencies (f [n]).

In the block 620 the frequency f _{[n] is} observed for the values of f I _{f [n]} and Q _{f [n]} .

In the block 630 I _{f [n] is} rotated to within a tolerance of a selected angle, such. B. zero degrees, to lie. In the block 640 Q _{f [n]} is rotated by exactly the same amount.

In one embodiment, it is more important to rotate I and Q identically than to rotate I perfectly to 0 degrees, or to some other selected angle. In one example, only exchanging real and imaginary expressions and reversing their signs can be used to rotate I to between 0 and 90 degrees (no multiplier required). Comparing the amplitude of the real and imaginary parts of I can be used to determine if I is between 0 and 45 or between 45 and 90 degrees. If the imaginary part of I is larger than the real part of I, if I is between 45 and 90 degrees, then there should be a rotation of -45 degrees, so that it is between 0 and 45 degrees. This can be done inaccurately with a multiplier of only a few bits, since a low gain error can be introduced without consequences, provided that this is done equally for I and Q. Considering that I is compared to Q, a gain error that both have is irrelevant. This technique can be further elaborated to rotate I conveniently sufficiently close to 0 degrees, thus gaining most of the benefit that the complete rotation of I can provide to 0 degrees or to another chosen angle.

The rotation of this block is used to cancel the phase noise. No useful information is lost through rotation because the correction is based on observing the relationship difference between I and Q. The relationship generally does not provide useful information (at least for that particular observation). Without the rotation, I and Q would each yield a near zero sum, it is believed that phase noise and low time domain averaging result in an equal distribution of the averaged phase of I and Q.

In the block 642 the rotated I and Q values are accumulated in the accumulator, which in block 602 was deleted. This can be a complex accumulator since both I and Q have a complex representation.

In the block 650 n is incrementally increased.

The decision block 660 Check if this is the last n. If this is not the case, the control moves back to the block 610 to evaluate the next test frequency. When the last n has been reached, the controller continues with block 670 continued.

In the block 670 For example, a means such as a mathematical mean for the accumulated I results can be calculated. A mathematical means is disclosed only as a possible example, and many other averaging, accumulating or aggregating methods may be used.

In the block 680 can a means such. For example, a mathematical mean for the accumulated Q results can be calculated.

In the block 690 this procedure is carried out.

The accumulated and averaged values of I and Q from the procedure 600 can be used to calculate IQ * / QQ *. This differs from the previously discussed techniques which accumulate IQ * separately from QQ *, and in the end shares IQ * with QQ *. In the process 600 I and Q must be rotated for each FFT result, but the multiplication of IQ * and QQ *, as well as the division of IQ * / QQ *, are done only once at the end.

The system and method of the preceding paragraphs realize numerous advantages.

Identifying the IQ imbalance free of obscuring errors is not only important to allow for accurate identification of the required correction, but also to allow the identification of a simplest implementation for this correction.

If the IQ imbalance is accurately isolated into concentrated analog sources (see error model), the complexity of the implied correction is significantly reduced.

Compensation of IQ imbalance would be more costly if attempted in a single correction stage because:
A relatively low order transfer function is sufficient to describe each individual concentrated source because each is associated with a finite number of dominant analog error sources. As soon as error sources combine, the apparent complexity of the correction increases significantly due to the fundamentally different influence that each concentrated source of error contributes to IQ imbalance, especially with respect to the positive versus negative frequency impact.

It is possible to achieve a very efficient and very accurate correction, provided that the concentrated sources of error are identified once and compensated in the opposite order for generation. This allows each concentrated error source to receive compensation from a correction structure that is optimized specifically for each situation, thereby enhancing both the correction accuracy and the computational efficiency.

However, if there is an observation error (especially if it is random), a correction with a significantly higher order is implied; the simpler underlying transfer function of the true IQ imbalance is hidden.

At least, an observation error that is not eliminated increases the order and complexity of the implied IQ imbalance correction and thus increases the strength and cost associated with this implementation for silicon.

If the observation noise is too high, a fixed hardware implementation may be sufficient to represent the intended transfer function. If this happens, the realized correction filter can show a significant transfer function, which ranks between directly specified frequency points. As a consequence, the IQ mismatch could be increased rather than reduced by the correction structure.

It should be noted that the activities explained above with reference to the figures may be applied to all integrated circuits including signal processing (eg, gesture signal processing), particularly those capable of executing specialized software programs or algorithms, of which some may be associated with the processing of digitized real-time data. Certain embodiments may relate to multi-DSP signal processing, floating point processing, signal / control processing, fixed function processing, microcontroller applications, etc. In certain contexts, the features discussed herein may be applied to medical systems, scientific instruments, wireless or wired communications, radar, industrial process control, audio and video equipment, current sampling, instrumentation (which can be highly accurate), and other digital processing-based systems. In addition, certain embodiments discussed herein may be provided in digital signal processing technologies for medical imaging, patient monitoring, medical instruments, and home health care. This could include lung monitors, accelerometers, heart rate monitors, pacemakers, etc. Other applications may include automotive technologies for safety systems (such as stability control systems, driver assistance systems, brake systems, infotainment, and internal applications of any kind). In addition, powertrain systems (eg, in hybrid and electric vehicles) can use products with high-precision data conversion in battery monitoring, control systems, report controls, maintenance activities, and more. In yet other example scenarios, the teachings of the present disclosure may be applied in the industrial markets that include processing control systems that help promote productivity, energy efficiency, and reliability. In consumer applications, the teachings of the signal processing circuits discussed above may be used for image processing, autofocus, and image stabilization (eg, for digital still cameras, camcorders, etc.). Other applications for consumers may include audio and video processors for Home theater systems, DVD recorders and HD televisions include. Still other applications for the consumer may include advanced touch screen controls (eg, for each type of portable media device). Thus, such technologies could easily make up parts of smartphones, tablets, security systems, personal computers, gaming technologies, virtual reality, simulation training, etc.

The foregoing addresses features of some embodiments so that those skilled in the art will better understand the aspects of the present disclosure. Those skilled in the art should understand that they may already use the present disclosure as a basis for the design or modification of other processes and structures to accomplish the same purposes and / or achieve the same benefits of the embodiments introduced herein. Those skilled in the art should also recognize that such equivalent structures are not to depart from the spirit and scope of the present disclosure and that many different changes, substitutions, and alterations can be made herein without departing from the spirit and scope of the present disclosure.

The particular embodiments of the present disclosure may easily include a package of a system-on-chip (SOC) central processing unit (CPU). An SOC represents an integrated circuit (IC) that integrates the components of a computer or other electronic system into a single chip. It can include digital, analog, mixed signal and radio frequency functions: all of these can be provided on a single chip substrate. Other embodiments may include a multi-chip module (MCM) having a plurality of chips disposed within a single electronic package and configured to tightly interact with each other through the electronic package. In various other embodiments, the digital signal processing functionality may be implemented in one or more silicon cores in application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), and other semiconductor chips.

In example implementations, at least some portions of the processing activities depicted herein may also be implemented in software. In some embodiments, one or more of these features may be implemented in hardware provided outside of the elements of the disclosed figures, or consolidated in a suitable manner to achieve the desired functionality. The various components may include software (or reciprocal software) that may coordinate to accomplish the operations outlined herein. In still other embodiments, these elements may include any suitable algorithms, hardware, software, components, modules, interfaces, or objects that simplify their operations.

In addition, some of the components associated with the described microprocessors may be removed or otherwise consolidated. In a general sense, the arrangements illustrated in the figures may be more logical in their representations, while a physical architecture may include various permutations, combinations and / or hybrids of these elements. It is important to note that countless possible design configurations can be used to achieve the operational goals described herein. Accordingly, the associated infrastructure has a myriad of replacement arrangements, design choices, device capabilities, hardware configurations, software implementations, equipment options, etc.

Any and suitably configured processor component may execute any type of instruction associated with the data to achieve the operations detailed herein. Any processor disclosed herein could transform an item or item (eg, data) from one state to the other state. In another example, some of the fixed logic or programmable logic activities depicted herein (eg, software and / or computer instructions executed by a processor) may be implemented, and the elements identified herein may be one of a kind programmable processor, a programmable digital logic (eg, a Field Programmable Gate Array (FPGA), an Erasable Programmable Read Only Memory (EPROM), an Electrically Erasable Programmable Read Only Memory (EEPROM), an ASIC that includes digital logic, Software, code, electronic instructions, flash memory, optical discs, CD-ROMs, DVD ROMs, magnetic or optical cards, other types of machine-readable media suitable for storing electronic instructions, or any combination thereof For example, processors may store information in any suitable type of non-volatile storage medium (e.g., Rando m Access Memory (RAM), Read Only Memory (ROM), Field Programmable Gate Array (FPGA), Erasable Programmable Read Only Memory (EPROM), Electrically Erasable Programmable ROM (EEPROM), etc.), Software, Hardware, or in any other suitable component, device, element or object, if appropriate and based on specific requirements. Further, the information traced, sent, received, or stored in a processor could be stored in any database, register, table, cache, series, control list, or memory structure based on particular requirements and implementations, all of which are referenced in an appropriate time frame could be provided. Each of the memory parts discussed herein should be constructed to be encompassed within the broad term "memory." Likewise, all of the potential processing elements, modules and machines described herein should be constructed to be encompassed within the broad term "microprocessor" or "processor". In addition, in various embodiments, the processors, memory, network cards, buses, memory devices, related peripherals, and other hardware elements described herein may be implemented by a processor, memory, and other related devices configured with software or firmware. to emulate or virtualize the functions of these hardware elements.

Computer program logic that implements all or part of the functionality described herein is implemented in various forms, including, but not limited to, a source code form, a computer-executable form, and various intermediate forms (e.g., forms that are derived from an assembler, a compiler , a linker or a locator). In one example, the source code includes a number of computer program instructions that are implemented in various programming languages, such as computer programs. For example, an object code, an assembly language, or a higher-level language such as OpenCL, Fortran, C, C ++, JAVA, or HTML for use with various operating systems or operating environments. The source code can define and use various data structures and communication messages. The source code may be in a computer-executable form (eg, via an interpreter), or the source code may be converted to a computer-executable form (eg, via a translator, assembler or compiler).

In the explanations of the above embodiments, the capacitors, buffers, graphic elements, interconnect boards, clocks, DDRs, camera sensors, splitters, inductors, resistors, amplifiers, switches, digital cores, transistors, and / or other components can be easily replaced, substituted, or otherwise be modified to accommodate certain requirements of the circuit. In addition, it should be noted that the use of complementary electronic devices, hardware, nonvolatile software, etc., provides an equally feasible option for implementing the teachings of the present disclosure.

In an exemplary embodiment, any number of electrical circuits of the FIGURES may be implemented on a board of an associated electronic device. The board may be a generic switchboard that can hold various components of the electronic device's internal electronic system and also provide connections for the other peripherals. In particular, the board may provide the electrical connections with which the other components of the system can communicate electrically. All suitable processors (including digital signal processors, microprocessors, supporting chip sets, etc.), memory elements, etc., may be suitably coupled to the board based on particular configuration requirements, processing requirements, computer designs, etc. Other components such. External memory, additional sensors, audio / video control, and peripheral devices can be attached to the board as plug-in cards, via cable, or integrated into the board itself. In another exemplary embodiment, the electrical circuits of the FIGURES may be implemented as stand-alone modules (eg, a device with associated components and a circuit configured to perform a specific application or function), or may be implemented as a plug-in. Modules may be implemented in the application specific hardware of the electronic devices.

It should be noted that the numerous examples provided herein describe the interaction with two, three, four or more electrical components. This was done only for the purpose of clarity and exemplification. It is understood that the system can be consolidated in any suitable way. According to similar design alternatives, all of the illustrated components, modules, and elements of the FIGURES may be combined in various possible configurations, all of which are well within the broad scope of this description. In certain cases, it may be easier to describe one or more of the functionalities of a given set of operations by referring only to a limited number of electrical elements. It should be understood that the electrical circuits of the FIGURES and the teachings thereof are readily scalable and capable of accommodating a large number of components as well as more sophisticated arrangements and configurations. Accordingly, the examples provided should not limit the scope or obstruct the ready teachings of the electrical circuits potentially applied to a myriad of other architectures.

Numerous other changes, substitutions, variations, and modifications may be made by those skilled in the art, and it is intended that the present disclosure encompass all such changes, substitutions, variations, and modifications as fall within the scope of the appended claims , To assist the United States Patent and Trademark Office (USPTO) and, in addition, all readers of each patent granted hereof in an interpretation of the appended claims, the Applicant wishes to note that Applicant: (a ) does not intend that any of the appended claims refer to paragraph six (6) of the 35 USC Section 112 as existing at the date of filing, unless the words "means to" or "step to" are specifically used in the specific claims; and (b) does not intend, by any statement in the description, to limit this disclosure in any way that is not otherwise reflected in the appended claims.

Claims (27)

An integrated circuit for correcting a quadrature error in a received signal comprising a circuit and logic adapted to: receive a separate error correction coefficient for each of a plurality of sequential error sources; and apply the error correction coefficients to the received signal in a reverse order of the sequence of error sources. An integrated circuit according to claim 1, wherein each error source is characterized by a concentrated error model. The integrated circuit of claim 1, wherein the plurality of error sources comprises a local oscillator error (LO error) and a baseband error (BB error). The integrated circuit of claim 3, wherein the plurality of error sources further comprises a predemodulation error (PD error). The integrated circuit of claim 4, wherein the sequence of error sources is PD, LO, BB, and wherein the sequence of applying the error correction is BB, LO, PD. The integrated circuit of claim 5, wherein calculating an error correction factor for each stage comprises: Generating a complete set of frequency domain BB correction coefficients; Generating a LO correction scalar; and Generation of complex filter coefficients. The integrated circuit of claim 6, wherein the circuit and logic are further configured to convert the LO correction scalar

to compensate for the compensation applied in the baseband correction filter (QFIR) for LO phase error induced amplitude loss compensation. The integrated circuit of claim 3, wherein the circuit and logic are further configured to convert a baseband correction filter gain

to compensate for LO phase error induced amplitude loss, where E _{LO is} the LO phase error. The integrated circuit of claim 3, wherein the LO phase error can be counteracted by the use of a configurable LO delay in an analog mixer. The integrated circuit of claim 9, further comprising a feedback loop to adjust the integrated circuit to compensate for non-linearities. The integrated circuit of claim 1, further comprising a plurality of coefficient buffers configured to support optimal coefficient update options. The integrated circuit of claim 1, further comprising a baseband processor, and wherein if the coefficient updates are not managed by the baseband processor, the generation of unwanted artifacts after the coefficient update can be prevented by expanding to new coefficients using a plurality of smaller steps. Quadrature receiver comprising: a receiver channel configured to receive a radio frequency (RF) signal; and a quadrature error correction that is designed to: receive a separate error correction coefficient for each of a plurality of sequential error sources; and apply the error correction coefficients to the RF signal in a reverse order to the sequence of error sources. A quadrature receiver according to claim 13, wherein each error source is characterized by a concentrated error model. The quadrature receiver of claim 13, wherein the plurality of error sources comprises a local oscillator error (LO error) and a baseband error (BB error). The quadrature receiver of claim 15, wherein the plurality of error sources further comprises a predemodulation error (PD error). The quadrature receiver of claim 16, wherein the sequence of error sources is PD, LO, BB, and wherein the sequence of applying the error correction is BB, LO, PD. The quadrature receiver of claim 15, wherein calculating an error correction factor for each stage comprises: Generating a complete set of frequency domain BB correction coefficients; Generating a LO correction scalar; and Generation of complex filter coefficients. The quadrature receiver of claim 18, further comprising scaling the LO correction scalar

to compensate for the compensation applied in the baseband correction filter (QFIR) for the LO phase error induced amplitude loss. The quadrature receiver of claim 15, further comprising scaling a baseband correction filter gain

to compensate for LO phase error induced amplitude loss, where E _{LO represents} an LO phase error. The quadrature receiver of claim 15, wherein the LO phase error can be counteracted using a configurable LO delay in an analog mixing console. The quadrature receiver of claim 21, further comprising a feedback loop to adjust the quadrature error correction to compensate for non-linearities. The quadrature receiver of claim 13, further comprising a plurality of coefficient buffers configured to support optimal coefficient update options. The quadrature receiver of claim 13, further comprising a baseband processor, and wherein if the coefficient updates are not managed by the baseband processor, the generation of unwanted artifacts after the coefficient update can be prevented by expanding to new coefficients using a plurality of smaller steps. A method of detecting a quadrature error in a received signal, comprising: Receiving a separate error correction coefficient for each of a plurality of sequential error sources; and Applying the error correction coefficients to the RF signal in a reverse order of the sequence of the error sources. The method of claim 25, wherein each source of error is characterized by a concentrated error model. The method of claim 25, wherein the plurality of error sources in order comprises a predemodulation error (PD error), a local oscillator error (LO error) and a baseband error (BB error), and wherein the sequence of applying the error correction BB, LO, PD is.

Images