WO2018224161A1

WO2018224161A1 - Reducing distortions in amplified signals to be radiated by an antenna

Info

Publication number: WO2018224161A1
Application number: PCT/EP2017/064105
Authority: WO
Inventors: Lei GUAN
Original assignee: Nokia Technologies Oy
Priority date: 2017-06-09
Filing date: 2017-06-09
Publication date: 2018-12-13

Abstract

Circuitry, a method, means and a computer program configured to reduce distortions in an amplified signal to be radiated by an antenna. The circuitry comprises: statistical property monitoring logic operable to monitor at least one statistical property of a digital input signal; and digital predistortion logic operable to select a predistortion function to apply to the input digital signal in dependence upon the monitored at least one statistical property of the digital input signal.

Description

REDU CIN G DISTORTION S IN AMPLIFIED SIGNALS TO BE RADIATED BY

AN ANTENNA

FIELD OF THE INVENTION

The invention relates to the field of systems for feeding signals to antenna.

BACKGROUND

Signals fed to one or more antenna are generally processed and amplified prior to being output. In a practical wireless system with physical RF transmitters, RF power amplifiers (PAs) are required to provide reasonable radiating power from each transmitter. Such amplifiers have a region where they operate linearly and a region beyond this where they do not. Operating in the non-linear region results in signals outside of the required bandwidth and in distortions in the signal. This results not only in the signal itself being difficult to decode, but also provides interference for neighbouring signals. Operating in the amplifier's linear region addresses this, but operation in this region is not efficient. Thus, these amplifiers if used in a region where they are efficient introduce nonlinear distortion into the signal causing both in-band signal quality degradation (which causes problems to the transmitter itself) and out-of- band spectrum regrowth (which cause problems to others transmitters working in the adjacent frequency band).

To address this problem two approaches are commonly used:

1) Backing-off approach : In this approach , radio frequency power amplifiers are operated in the linear operating region, i.e. backing-off the power from the saturation operating region. A drawback of this conventional approach is that it leads to very low power efficiency.

2) Digital predistortion (DPD) approach : This involves the input signal being pre- distorted to compensate for distortions that will arise at the amplifier, such that the amplified signal has reduced distortions. Figure 1A shows a conventional DPD-enabled RF transmitter with two signal paths, a forward data path and a feedback data path. The amplified radio frequency signal is sampled and fed back via sampling receiver devices. The attenuator is used for reducing the power of the feedback signal, a down- converter converts the signal from a high frequency radio signal to a lower frequency signal, and an analogue to digital converter returns the signal to a digital signal. This digital signal is then compared with the signal that generated it at the digital signal processing circuit and from this comparison the pre-distortion function applied to the data signal on the forward data path is updated if required to mitigate for any differences detected between the two signals. Due to its satisfactory linearization performance and flexibility, DPD has been largely used as a preferred option to reduce the nonlinear distortion introduced by the RF PAs when driving RF PAs into nonlinear saturation region .

However, directly applying this DPD architecture for multiple RF transmitters based system (such as Massive MIMO system) results in a system that is expensive both in hardware and in power. Figure IB shows such a device.

It would be desirable to be able to address the non-linear effects of an amplifier in a transmitting system without unduly increasing the costs or power consumption of such a system.

SUMMARY

A first aspect of the present invention provides circuitry operable to reduce distortions in an amplified signal to be radiated by an antenna, said circuitry comprising: statistical property monitoring logic operable to monitor at least one statistical property of a digital input signal; digital predistortion logic operable to select a predistortion function to apply to said input signal in dependence upon said monitored at least one statistical property of said digital input signal.

One way of dealing with distortions produced by non-linearities in amplifiers in a system feeding an antenna is to use a predistortion function to predistort the signal prior to it being amplified. The predistortion acts to cancel out or at least reduce the distortion generated by the amplifier. Conventionally this predistortion function has been generated using a feedback path the feedback path allowing for real time updating of the pre-distortion function to account for changes in the input signal and changes in the operating characteristics of the amplifier. This continual feedback can be expensive on both processing power and hardware, particularly for multiple antenna systems such as massive MIMO and for wideband signals, where sampling rates are high. The inventor recognised that the statistical properties of signals and in particular, factors such as PAPR (peak to average power ratio) affect the distortion that the signal experiences at the amplifier. Thus, by monitoring the statistical properties of a digital signal input to a transmitting system an indication of a suitable predistortion function can be obtained, this can be used in the selection of such a function obviating or at least reducing the need for feedback signals.

The inventor recognised that selecting the predistortion function based on the statistical properties of the input signal allows for a more efficient updating of the predistortion function to be performed without the requirement for continual feedback. In particular as the digital predistortion functions share similar statistical properties to the transmission signals, the DPD functions for signals with similar statistical properties will be valid and provide satisfactory performance.

Furthermore, in some systems where statistical properties of a signal remain fairly constant for long periods of time, this system is particularly efficient. For example, during precoding in massive MIMO systems, precoding for beamforming is performed and this precoding affects the statistical properties of the signals. For some scenarios such as the transmission of signals to stationary nodes, the beamforming or precoding is relatively constant for prolonged periods of time, meaning that the predistortion functions do not need to be updated regularly. Embodiments provide particularly effective low processing predistortion in such cases. Although the digital predistortion logic may be operable to select different predistortion models to apply to the input signal, in some embodiments said circuitry further comprises a data store operable to store a plurality of sets of digital predistortion coefficients for said predistortion functions, each set being stored linked to a statistical property, said digital predistortion logic being configured to select one of said sets of digital predistortion coefficients in dependence upon a value of said monitored statistical property.

In some embodiments, the predistortion model applied to the signal may not change but the coefficients applied to that model may vary as the statistical properties of the input signal varies. In such a case, an efficient way of updating the predistortion function is to have a data store that stores the predistortion coefficients along with an indication of the statistical property of a signal to which they should be applied. Thus, monitoring the statistical property of the input signal allows the appropriate set of coefficients to be selected from the data store and the corresponding predistortion function to be applied to the input digital signal. Although embodiments can be applied to input digital signals of any bandwidth, they are particularly advantageous for wider band signals, for example, for signals having a bandwidth greater than 100MHz or even in some cases greater than one to several thousand MHZ. Wider band signals pose particular problems for generating predistortion functions using feedback as the sampling of such wide band signals can be expensive. Eliminating or at least reducing the need for such feedback signals can therefore be particularly advantageous in such cases.

In some embodiments, said statistical monitoring logic is configured to monitor said input digital signals periodically and in response to determining a change in a monitored statistical property said digital predistortion logic is operable to select an updated predistortion function to apply to said input signal.

As noted previously, the predistortion function that is appropriate to a particular input signal is dependent on the statistical properties of that signal. Thus, when a change in a monitored statistical property is determined, for example when it meets or exceeds a threshold value, then this is an indication that a different predistortion function may be more appropriate and thus the predistortion logic can be operable in response to detecting this to select an updated predistortion function and apply that to the input signal. In this regard, where the predistortion function coefficients are stored in a table indexed by statistical properties, then the statistical properties of the input signal being detected as moving from one index value to a further index value may be the indication that different coefficients should be used. In some embodiments, said circuitry further comprises a signal sampler; a feedback path for routing said amplified signal to said signal sampler; predistortion coefficient generating logic configured to receive said sampled signal and said digital input signal corresponding to said sampled signal and to generate predistortion coefficients from a comparison of said signals; a data store configured to store said generated coefficients along with an indicator of a statistical property of said input digital signal, said statistical property being received from said statistical property monitoring logic.

In some embodiments, the predistortion coefficients that are stored and later selected are generated by the circuitry using a feedback path. The predistortion coefficient generating logic compares the fed back signal with the input signal corresponding to it and generating predistortion coefficients from a comparison of the signals. These generated coefficients are stored along with an indicator of a statistical property of the input digital signal and can later be selected for use by the digital predistortion logic.

In some embodiments, said signal sampler comprises an analogue to digital signal convertor.

In some embodiments, said signal sampler is configured to sample said amplified signal at a sampling rate that is lower than a sampling rate of said input digital signal. One advantage of using the statistical properties of the digital signal and it being changes in these properties that are of interest, is that the analysis does not focus on individual samples in a time or frequency domain. This allows sampling at a lower sampling rate to be performed and provide acceptable results. This is important as high sampling rates are expensive in hardware and power and thus, allowing a lower sampling rate, can significantly reduce the cost of the hardware on the feedback loops.

In some embodiments, said sampling rate is lower than a Nyquist sampling rate for said amplified signal. In some embodiments, said comparison comprises performing convolution based peak detection to determine correlation levels between samples of said feedback signal and said digital input signals, a time offset between samples providing a highest peak of correlation being a time offset for shifting one set of signals to provide aligned pairs for comparison.

As the sampling rate of the fed back signal is different to the sampling rate of the input digital signal, the actual samples are not aligned in time and prior to any comparison alignment may be required. This can be done by performing convolution based peak detection to determine correlation levels between the samples and when the most correlated signals have been determined, a time offset between these samples can be determined which provides a time offset for shifting one set of signals to provide aligned pairs for comparison.

Conventionally when sampling a signal the sampling rate is related to the data rate, and is conventionally set at three times the data rate of the signal. In the current case, as it is the statistical properties of the sampled fed back signal that are of interest, the sampling rate can be made to be lower than the data rate and in some cases

significantly lower.

Although the comparison can be done in a number of ways, in some embodiments, said comparison comprises a muti-rate least squares identification.

In some embodiments, said predistortion coefficient generating logic is configured to generate said coefficients for input digital signals with different statistical properties.

The predistortion coefficients may be generated during an initial training phase. As these coefficients change with statistical properties, the signals input during this training phase may be signals with different statistical properties. The predistortion coefficient generating logic will generate different coefficients by responding to the signals with different statistical properties. These coefficients will be stored with an indication of the statistical properties to which they relate. Appropriate coefficients can then be selected during normal operation in dependence upon the statistical properties of the digital input signal.

In some embodiments, said predistortion coefficient generating logic is configured to generate a lookup table indexed by statistical property for said coefficients during an initial training phase.

Although the data store storing the predistortion coefficients may have a number of forms, in some embodiments it comprises a look-up table that is indexed by statistical properties such that the relevant coefficients can be read out depending on the statistical properties of the input digital signal.

In some embodiments, said predistortion coefficient generating logic is configured to periodically update said predistortion coefficients during operation.

In addition to their initial generation during a training phase, the coefficients may be updated. This can be done periodically during operation.

Alternatively and/ or additionally, the coefficients may be updated on demand in response to detecting that the predistortion logic is no longer operating to a particular standard. In order to determine this, the circuitry further comprises monitoring circuitry operable to monitor said amplified signal and in response to detecting a portion of said signal outside a predetermined bandwidth, said circuitry is configured to control said predistortion coefficient generating logic to update said predistortion coefficients.

One result of the predistortion functions not working particularly well is that out of bound signals will appear in the output signals. Monitoring of these out of bound signals is one way of determining that the predistortion functions applied to the digital input signals are no longer applying particularly effective predistortions and thus it is an indication that these should be updated.

In some embodiments, said circuitry is configured to reduce distortions in a plurality of amplified signals to be radiated by a plurality of antennas, said circuitry comprising: a plurality of statistical property monitoring logic modules each operable to monitor at least one statistical property of a corresponding one of a plurality of digital input signals; a plurality of digital predistortion logic modules each operable to select a predistortion function to apply to a corresponding one of said plurality of digital input signals in dependence upon said monitored at least one statistical property of said digital input signal.

Although the circuitry can be applied to a single amplified signal being sent to a single antenna, it is particularly advantageous when applied to a plurality of signals sent to a plurality of antenna. As noted previously, embodiments of the invention provide reduced costs in hardware and processing and thus where these costs are multiplied by having multiple signals, these savings are themselves multiplied.

In some embodiments, said circuitry further comprises a plurality of feedback paths for routing each of said plurality of amplified signal to at least one signal sampler; at least one predistortion coefficient generating logic module configured to receive said sampled signals and said digital input signals corresponding to said sampled signals and to generate said predistortion coefficients from a comparison of said signals; at least one data store for storing said generated coefficients for each of said plurality of predistortion logic modules along with an indicator of a statistical property of said corresponding input digital signal.

In some embodiments, said circuitry comprises one signal sampler, one predistortion coefficient generating logic module and one data store, said circuitry further comprising: switching circuitry for routing a selected one of said amplified signals to said signal sampler; and further switching circuitry for routing a corresponding selected one of said input digital signals to said predistortion coefficient generating logic module. Where there are multiple antenna and multiple feedback possibilities, in some embodiments there may simply be one predistortion coefficient generating logic module and one data store and one signal sampler with switching circuitry for routing a selected amplified signal to be fed back at any one time. In this regard, as

embodiments do not require continual feedback of the signals to determine the preferred predistortion function, it is acceptable for different signals to be fed back at different times, the hardware being shared on a time basis between the different signals.

In other embodiments, said circuitry comprises a plurality of signal samplers; a plurality of predistortion coefficient generating logic modules each configured to receive a corresponding one of said sampled signals and said digital input signal corresponding to said sampled signal and to generate said predistortion coefficients from a comparison of said signals; and a plurality of data stores for storing said generated coefficients for each of said plurality of predistortion logic modules along with an indicator of a statistical property of said input digital signals, said statistical property being received from said statistical property monitoring logic.

Alternatively, there may be a feedback loop for each signal path with each feedback loop having its own signal sample, predistortion coefficient generating logic module and data store. In this regard, owing to embodiments allowing a significantly reduced sampling rate for the fed back signal, the hardware required for the feedback loops is significantly less expensive than in conventional systems and as such, providing multiple feedback paths with multiple hardware may be acceptable. In some embodiments, said plurality of antenna comprise a massive MIMO antenna system.

As mentioned earlier, embodiments are particularly advantageous when applied to multiple antenna systems and in particular to massive MIMO antenna systems where there are many antenna, in some cases 64 or more. A second aspect of the present invention provides a method for reducing distortions in an amplified signal to be radiated by an antenna, said method comprising: monitoring at least one statistical property of a digital input signal; selecting a predistortion function to apply to said input digital signal in dependence upon said monitored at least one statistical property of said digital input signal; applying said selected predistortion function to said input digital signal.

In some embodiments, said step of selecting said predistortion function comprises selecting a set of digital predistortion coefficients for a predistortion model from a data store, each set being stored in said data store linked to a statistical property.

In some embodiments, said digital input signal comprises a wideband signal having a bandwidth greater than 100MHZ. In some embodiments, said step of monitoring is performed periodically and in response to determining a change in a monitored statistical property, said selecting step selecting an updated predistortion function to apply to said input signal.

In some embodiments, said method further comprises: routing said amplified signal to a signal sampler on a feedback path ; receiving said sampled signal and said digital input signal corresponding to said sampled signal and generating predistortion coefficients from a comparison of said signals; storing said generated coefficients along with an indicator of a statistical property of said input digital signal, said statistical property being received from said statistical property monitoring logic.

In some embodiments, said step of sampling further comprises converting an analogue signal to a digital signal.

In some embodiments, said step of sampling comprises sampling said amplified signal at a sampling rate that is lower than a sampling rate of said input digital signal.

In some embodiments, said comparison comprises a multi-rate least squares identification.

In some embodiments, said step of generating said predistortion coefficients is performed for input digital signals with different statistical properties. In some embodiments, said step of generation said predistortion coefficients comprises generating a lookup table indexed by statistical property for said coefficients during an initial training phase.

In some embodiments, said step of generating predistortion coefficients is performed periodically during operation.

In some embodiments, said method further comprises monitoring said amplified signal and in response to detecting a portion of said signal outside a predetermined bandwidth updating said predistortion coefficients.

In some embodiments, said method is for reducing distortions in a plurality of amplified signals to be radiated by a plurality of antennas, said method comprising: monitoring at least one statistical property of a plurality of digital input signals;

selecting a predistortion function to apply to a corresponding one of said plurality of digital input signals in dependence upon said monitored at least one statistical property of said digital input signal.

A third aspect of the present invention provides, a computer program which when executed by a computer is operable to control said computer to perform a method according to a second aspect of the present invention.

A fourth aspect of the present invention provides a means for reducing distortions in an amplified signal to be radiated by an antenna, said means comprising: statistical property monitoring means for monitoring at least one statistical property of a digital input signal; digital predistortion means for selecting a predistortion function to apply to said input digital signal in dependence upon said monitored at least one statistical property of said digital input signal. In some embodiments, said means further comprises: a signal sampling means for sampling said amplified signal; a feedback path for routing said amplified signal to said signal sampling means; predistortion coefficient generating means for receiving said sampled signal and said digital input signal corresponding to said sampled signal and for generating predistortion coefficients from a comparison of said signals; a data storage means for storing said generated coefficients along with an indicator of a statistical property of said input digital signal, said statistical property being received from said statistical property monitoring logic.

The various means for may comprise software or circuitry configured to perform these functions.

Further particular and preferred aspects are set out in the accompanying independent and dependent claims. Features of the dependent claims may be combined with features of the independent claims as appropriate, and in combinations other than those explicitly set out in the claims.

Where an apparatus feature is described as being operable to provide a function , it will be appreciated that this includes an apparatus feature which provides that function or which is adapted or configured to provide that function.

BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the present invention will now be described further, with reference to the accompanying drawings, in which :

Figure 1A illustrates a digital predistorter and feedback loop according to the prior art; Figure IB illustrates multiple digital predistorters and feedback loops for a massive MIMO system according to the prior art;

Figure 2 illustrates a digital predistorter according to an embodiment;

Figure 3 illustrates a digital predistorter with feedback loop according to a further embodiment;

Figure 4 illustrates a digital predistorter for a massive MIMO system according to an embodiment; and

Figure 5 shows a flow diagram illustrating steps in a method according to an embodiment. DESCRIPTION OF THE EMBODIMENTS

Before discussing the embodiments in any more detail, first an overview will be provided. A system that provides digital predistortion DPD functions that are selected from a stored set of DPD functions based on statistical properties of the dgitial signal to be predistorted is provided. It is recognized that the DPD function and in particular DPD coefficients used in a DPD model will change with the statistical properties of the input signal. Thus, monitoring these statistical properties of the input signal and selecting different DPD coefficients as the statistical properties change is a way of selecting and updating DPD functions without a continuous feedback signal.

Figure 2 shows an example of such a system . A digital input signal 10 is received and monitored by a statistical properties monitoring module 20. DPD coefficients are selected from data store 30 in response to the detected statistical properties and these coefficients are applied to the model in digital predistorter 40 , such that the DPD function is updated as the statistical properties change.

Once the signal has been predistorted, the signal is then converted to a radio frequency signal prior to being amplified by amplifier 50 and output to antenna 60.

In order to populate the data store with appropriate DPD coefficients, an initial training stage may be performed where different data streams with different statistical properties (PAPR values and RMS power) are input to the system , a feedback loop determines appropriate coefficients for each of the different statistical properties and then stores them in memory 30 along with an indicator indicating the statistical properties to which they correspond.

Figure 3 shows the circuitry of Figure 2 with the additional feedback loop for populating and for updating the DPD coefficient pool. In this embodiment a coupler at the output to the amplifier 50 feeds the signal back to a sampler 70 which in this embodiment is an analogue to digital converter ADC. A signal from earlier in the forward path is also sampled at a different sampling rate and processing circuitry 80 correlates the signals and then compares the two signals that it determines correspond to each other and determines from this comparison DPD coefficients for the coefficient pool for the particular input signal currently being monitored. The statistical properties of this input signal are also fed from SPM module 20 such that the DPD coefficients that are determined can be stored associated with the statistical properties of the signal. During the initial training phase input signals with different statistical properties may be used so that the coefficients for a range of different statistical properties can be determined and stored.

Then during normal operation the statistical properties monitoring module 20 consistently monitors the statistical properties of the input data stream, and then according to different statistical properties (PAPR value and RMS value), a set of DPD coefficients will be selected from the data store 30 and put in place in the digital predistorters 40 to predistort the data streams.

The DPD coefficients pool can be updated by re-doing the training session if necessary. This can be done periodically or on demand in response to detecting distortions in the amplified signal. These can be detected by detecting out of band signals for example.

Conventionally in order to properly capture the output signals from amplifier 50 , usually a bandwidth of 3 to 5 times the bandwidth of the original input signal is used. For example, if the forward data path is transmitting an LTE signal with a 20MHz bandwidth (within some frequency band across 60MHz to lOOMHz bandwidth), conventionally to fully capture the 60MHz to lOOMHz bandwidth signal (depending on the requirement of interested linearization bandwidth) approximately 120Msps to 200Msps sampling rate Nyquist ADCs are required. For Marco basestations having a few RF transmitters, the requirement of the extra high-performance ADCs required for this sampling might be acceptable in view of the overall consideration of cost, total power consumption and corresponding linearization performance. However for systems with more transmitter chains such as Massive MIMO basestation applications, with many more transceivers (the number may above 64), it is quite clear that adding such high-performance DPD feedback paths will increase the overall cost of the system dramatically. Furthermore, for wide bandwidth applications the requirement of the sampling rate may also prove prohibitive.

Thus, although embodiments function for single data streams, they are particularly effective for multiple antenna systems such as Massive MIMO (Multiple Input Multiple Output) wireless systems. The basic concept of massive MIMO is to provide wireless communication services by utilizing a large number of antennas with corresponding transmitters to service multiple users within the same frequency band. In this way, the network capacity and equivalent data throughput can be significantly increased. In order to provide reasonable coverage by the Massive MIMO basestations, the output RF power of each formed antenna beam needs to be high enough. In other words, an RF power amplifier (PA) on each of the transmitter chain is needed. The power efficiency of RF power amplifiers will significantly affect the overall system

performance regarding the energy consumption, so it is required to drive the RF power amplifiers into high-efficient operation region. However, the RF power amplifiers are inherently nonlinear at high-efficient operation region, in other words, RF power amplifiers introduce nonlinear distortion into the signal causing both in-band signal quality degradation (cause problems to the transmitter itself) and out-of-band spectrum regrowth (cause problems to others transmitters working in the adjacent frequency band). For handling the out-of-band spectrum issue in the Massive MIMO basestations, RF engineers back-off the power to let the RF power amplifiers work in the linear operation region. However, this back-off approach leads to very low power efficiency, for example < 15% power added efficiency (PAE) at the final power amplifier stage.

In short, it would be desirable to reduce the out-of-band spectrum regrowth (as well as in-band distortion) and improve the PA efficiency of each individual transmitter in the Massive MIMO basestations.

Embodiments reduce the cost and power consumption of DPD-based PA linearization for the Massive MIMO application, by jointly utilizing an advanced signal processing algorithm and compact hardware architecture. A statistics-stationary DPD approach that typically includes two procedures is used: 1) DPD coefficient training, 2) real-time signal predistortion. L. Guan and A. Zhu, "Optimized Low-Complexity Implementation of Least Squares-Based Model Extraction for Digital Predistortion of RF Power Amplifiers", IEEE Transactions on Microwave Theory and Techniques, Vol. 60 , no. 3 , pp. 594-603, Mar. 2012 looked at least-squares - based parameter extraction when determining DPD coefficients.

In the present technique the statistical properties refer to for example, the peak to average power ratio (PAPR) and average normalized power or root mean square (RMS) power. It has been concluded that as long as the training signal used for deriving the DPD coefficients share similar statistical properties to the transmission signal, the DPD coefficients will be valid and provide satisfactory performance. In the Massive MIMO base-stations, there is a requirement to generate multiple space division multiplexed data streams via a so-called precoding approach. Although the precoding is mainly done in the frequency domain at the baseband processing unit, it is equivalent to apply asymmetrical finite impulse response (FIR) filter to each input data stream to output data stream (toward antennas). Therefore, we can say, the precoding approach will affect the statistical properties of the signals, and the DPD performance will be affected correspondingly. Our solution is to build a DPD coefficients LUT (look up table) pool in a Massive MIMO system that will be stored in memory and the DPD coefficient LUT pool will be indexed by the estimation of the statistical properties of each data stream to be transmitted by the Massive MIMO transmitter.

Secondly, in order to reduce the hardware cost for DPD feedback paths at each transmitter, we introduce an approach named as shadow sampling identification, which allows us to capture the PA feedback signal at a sampling rate lower-than the Nyquist sampling rate and perform system identification (DPD coefficients training). This approach can significantly reduce the requirement of sampling rate at the DPD feedback path, enabling low-cost ADC in-the-loop DPD approach.

The proposed solution is demonstrated in Figure 4. Figure 4 shows six modules:

1) SPM (statistical properties monitor), this module gathers the statistical

information of the transmission data streams. The typical statistical information used here are PAPR value and RMS power value. This module is re-configurable for different monitoring periods, which will be required for different applications, such as fast DPD coefficients updating for bursty traffic or slow DPD coefficients updating.

Digital Switch : this is a simple multiplexer in any digital logic platform, e.g., FPGA. This switch is used to feed different data streams together with statistical properties to the DPD coefficients training module.

3) Analog Switch : this is used to select the PA output from different transmitters to be fed to the DPD coefficients training module.

4) Low-profile ADC based compact RF to Digital Chain : this module contains several sub-modules, the main modules include:

a a compact band-pass filter (BFP) to select the interested linearization band, e.g., 80MHz at 2.6GHz;

b a digital control step attenuator to adjust the feedback power properly to the ADC comfortable zone to maximize the performance of data acquisition ; c. Frequency mixer: to down-convert the RF signal to the baseband or intermediate frequency (IF).

d. low-profile ADC, for example 50MSPS sampling rate low power ADC with input bandwidth of 200MHz. Due to the introduction of the shadow sampling approach, the aliasing introduced by the

undersampled ADC will not affect the DPD coefficient training procedure as long as the analog switching input bandwidth of the ADC can cover the bandwidth required. Both of the zero-IF two ADCs architecture or high-IF one ADC architecture can be used to convert the analog signals to the digital baseband signal. The ratio between high- data-rate (or high sampling rate) and low-data-rate (or low sampling rate) signals can be integer or not, but for simplicity, we recommend the integral relationship between high-data-rate forward signal and low- data-rate aliased PA feedback signal. For example, if the forward data path signal is running at 122.88MSPS, we can utilize the ADC with sampling rate of one quarter, e.g., 30.72MSPS or even one eighth, e.g., 15.36MSPS.

Shadow sampling based DPD coefficient training: this module will utilize both of the high-data-rate forward signal and low-data-rate aliased PA feedback signal to derive the DPD coefficients in a least squares type global estimation approach. The algorithm includes multi-rate based time alignment and multi- rate least squares identification.

a. Multi-rate time alignment algorithm : given high-data-rate forward signal XHR, and low-date-rate PA feedback signal yLR, since the sampling rate ratio between XHR and yLR is integer, say K, we can actually downsample the high-data-rate signal by a factor of K directly without an anti-aliasing filter. This will provide us with a group of K low-data- rate signals XLR, ¾ (k = 0 , 1, 2, .. K- l) with different phase offsets. Then, convolution based peak detection will be utilized to derive the correlation levels between XLR, ¾ and y_LR, the location of the highest peak of correlation result indicates the relative time-domain offset between XLR. k and yLR. We then can derive an aligned pair signal XHR and yLR by shifting either of the signal with estimated offsets. We name this sampling processing as shadow sampling, because the actual PA feedback signal is captured in an aliasing way (compared to the conventional non-aliasing way), and in the frequency domain the main carrier will be shadowed by its aliasing part. Multi-rate least squares identification algorithm : given a DPD model with M terms (including nonlinear distortion terms, linear memory terms, nonlinear memory terms), a multi-rate system transfer function can be achieved by selecting rows of the original over-determined equations ((K- l)*n equations) to form a number reduced over- determined equation (n equations), which row corresponds to the time aligned low sampling rate feedback samples. For example, let _m represents the corresponding m^th basis term of a DPD model (with memory length of /), the original over-determined equation can be formed as below.

where A_m = f_m (x_{l+ i},x_{2+ i},- ' -*Z+lxi ) i = 0, 1, ..., n-l, m = l, 2, . .,

Where c denotes the DPD coefficients. In the above equation, we assume we obtained the PA feedback signal y i in the high date rate case, however, in we may not have this high data rate feedback signal to perform DPD coefficient estimation (due to the introduction of shadow sampling). In this case the multi-rate over-determined equation via shadow sampling can be built as below.

where B' = f m ( ν^Λ1+(ί-1)χί ' ^Λ2+(Κ-ΐ)χί ' ' ^■ , X 1,+(Κ-ΐ)χ t ) i = 0, 1, ..., n-l, m = 1, 2, ...,

Here vector yi

L " ' " ^{l+ K l>xl} " ^{l+(K 1)x}" ^J corresponds to the low data captured PA feedback signal yLR, and this over-determined equation can be solved by least squares- based global optimization. The algorithm can be processed in a programmable device or system on chip (SoC) device.

6) DPD coefficients pool: the DPD coefficients are pre-calculated according to different possible statistical property combinations, and stored in a digital memory. The coefficients can be retrieved by indexing by the statistical property values. In order to maintain the power of the PA the same before and after DPD function, we need to regulate the RMS power in the digital domain. In other words, PAs are expecting the same RMS value input data stream before and after DPD approach. In this case the only variable required to index the DPD coefficient pool will be the PAPR value of the data stream. The size of the coefficient pool depends on the number of transmission chains, the DPD models and the resolution used for the statistical properties.

The main operation procedures are outlined below:

1) Step 1, DPD coefficient training stage. Before putting the system into normal operation, we can performance DPD training for each of the Massive MIMO

transmitter using different data streams with different statistical properties (PAPR values and RMS power), coefficients derived in this way are stored in a memory.

2) Step 2, DPD running stage. During normal operation , the SPM module consistently monitors the statistical properties of data streams, and then according to different statistical properties (PAPR value and RMS value), a set of DPD coefficients will be picked up (looked-up) and put in place in the digital predistorters to predistort the data streams.

3) The DPD coefficients pool can be updated by re-do the training session if necessary

The new solution proposed reduces the hardware requirement for the DPD feedback path, which leads to very efficient, low power consumption system architecture. The corresponding processing algorithms enable a very compact DPD-based linearization solution even for Massive MIMO base-stations. The conventional multiple high- performance DPD feedback paths may be replaced by a single low-profile ADC based compact feedback path, the complexity of the hardware is thereby dramatically reduced.

Figure 5 shows a flow diagram illustrating steps in a method according to an

embodiment. The statistical properties of an input digital data signal being fed towards an antenna are monitored and a predistortion function is selected to apply to this signal based on the monitored one or more statistical properties. The selected predistortion function is applied to the input signal and it is converted to an RF analogue signal, amplified and output at an antenna.

The amplified signal is monitored to determine if it is within a predetermined frequency band within acceptable limits. If it is found to be outside of this frequency band then this is an indication that the predistortion function is no longer operating well and should be updated. Thus, an update routine is initiated whereby the amplified signal is sampled on a feedback path and compared with the input signal from which it stemmed. The two signals may be correlated and a calculated time offset applied to one to align them prior to the comparison. Predistortion coefficients are then generated from the comparison of the signals and the coefficient data store is updated. In this regard a training input signal may be used for the update routine, this training signal having varying statistical properties, such that coefficients for different statistical properties can be determined and stored in the data store along with an indication of the statistical properties of the signal to which they relate. A person of skill in the art would readily recognize that steps of various above- described methods can be performed by programmed computers. Herein, some embodiments are also intended to cover program storage devices, e.g., digital data storage media, which are machine or computer readable and encode machine- executable or computer-executable programs of instructions, wherein said instructions perform some or all of the steps of said above-described methods. The program storage devices may be, e.g., digital memories, magnetic storage media such as a magnetic disks and magnetic tapes, hard drives, or optically readable digital data storage media. The embodiments are also intended to cover computers programmed to perform said steps of the above-described methods.

The functions of the various elements shown in the Figures, including any functional blocks labelled as "processors" or "logic", may be provided through the use of dedicated hardware as well as hardware capable of executing software in association with appropriate software. When provided by a processor, the functions may be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which may be shared. Moreover, explicit use of the term

"processor" or "controller" or "logic" should not be construed to refer exclusively to hardware capable of executing software, and may implicitly include, without limitation, digital signal processor (DSP) hardware, network processor, application specific integrated circuit (ASIC), field programmable gate array (FPGA), read only memory (ROM) for storing software, random access memory (RAM), and non-volatile storage. Other hardware, conventional and/ or custom, may also be included. Similarly, any switches shown in the Figures are conceptual only. Their function may be carried out through the operation of program logic, through dedicated logic, through the interaction of program control and dedicated logic, or even manually, the particular technique being selectable by the implementer as more specifically understood from the context. It should be appreciated by those skilled in the art that any block diagrams herein represent conceptual views of illustrative circuitry embodying the principles of the invention. Similarly, it will be appreciated that any flow charts, flow diagrams, state transition diagrams, pseudo code, and the like represent various processes which may be substantially represented in computer readable medium and so executed by a computer or processor, whether or not such computer or processor is explicitly shown.

The description and drawings merely illustrate the principles of the invention. It will thus be appreciated that those skilled in the art will be able to devise various arrangements that, although not explicitly described or shown herein, embody the principles of the invention and are included within its spirit and scope. Furthermore, all examples recited herein are principally intended expressly to be only for pedagogical purposes to aid the reader in understanding the principles of the invention and the concepts contributed by the inventor(s) to furthering the art, and are to be construed as being without limitation to such specifically recited examples and conditions.

Moreover, all statements herein reciting principles, aspects, and embodiments of the invention, as well as specific examples thereof, are intended to encompass equivalents thereof.

Claims

1. Circuitry operable to reduce distortions in an amplified signal to be radiated by an antenna, said circuitry comprising:

statistical property monitoring logic operable to monitor at least one statistical property of a digital input signal; and

digital predistortion logic operable to select a predistortion function to apply to said input digital signal in dependence upon said monitored at least one statistical property of said digital input signal.

2. Circuitry according to claim 1, said circuitry further comprising a data store operable to store a plurality of sets of digital predistortion coefficients for said predistortion functions, each set being stored linked to a statistical property, said digital predistortion logic being configured to select one of said sets of digital predistortion coefficients in dependence upon a value of said monitored statistical property.

3. Circuitry according to any preceding claim , wherein said digital input signal comprises a wideband signal having a bandwidth greater than 100MHZ.

4. Circuitry according to any preceding claim, said statistical property monitoring logic being configured to monitor said input digital signals periodically and in response to determining a change in a monitored statistical property said digital predistortion logic being operable to select an updated predistortion function to apply to said input signal.

5. Circuitry according to any preceding claim, said circuitry further comprising: a signal sampler;

a feedback path for routing said amplified signal to said signal sampler;

predistortion coefficient generating logic configured to receive said sampled signal and said digital input signal corresponding to said sampled signal and to generate predistortion coefficients from a comparison of said signals;

a data store configured to store said generated coefficients along with an indicator of a statistical property of said input digital signal, said statistical property being received from said statistical property monitoring logic.

6. Circuitry according to claim 5, said signal sampler comprising an analogue to digital signal converter.

7. Circuitry according to claim 5 or 6, wherein said signal sampler is configured to sample said amplified signal at a sampling rate that is lower than a sampling rate of said input digital signal.

8. Circuitry according to claim 7, wherein said sampling rate is lower than a Nyquist sampling rate for said amplified signal.

9. Circuitry according to claim 7 or 8 , wherein said comparison comprises performing convolution based peak detection to determine correlation levels between samples of said feedback signal and said digital input signals, a time offset between samples providing a highest peak of correlation being a time offset for shifting one set of signals to provide aligned pairs for comparison.

10. Circuitry according to any one of claims 5 to 9 , wherein said comparison comprises a multi-rate least squares identification.

11. Circuitry according to any one of claims 5 to 10 , wherein said predistortion coefficient generating logic is configured to generate said coefficients for input digital signals with different statistical properties.

12. Circuitry according to any one of claims 5 to 11, wherein said predistortion coefficient generating logic is configured to generate a lookup table indexed by statistical property for said coefficients during an initial training phase.

13. Circuitry according to any one of claims 5 to 12, wherein said predistortion coefficient generating logic is configured to periodically update said predistortion coefficients during operation.

14. Circuitry according to any one of claims 5 to 13 , said circuitry further comprising monitoring circuitry operable to monitor said amplified signal and in response to detecting a portion of said signal outside a predetermined bandwidth said circuitry is configured to control said predistortion coefficient generating logic to update said predistortion coefficients.

15. Circuitry according to any preceding claim, wherein said circuitry is configured to reduce distortions in a plurality of amplified signals to be radiated by a plurality of antennas, said circuitry comprising:

a plurality of statistical property monitoring logic modules each comprising said statistical property monitoring logic operable to monitor at least one statistical property of a corresponding one of a plurality of digital input signals;

a plurality of digital predistortion logic modules each comprising said digital predistortion logic each operable to select a predistortion function to apply to a corresponding one of said plurality of digital input signals in dependence upon said monitored at least one statistical property of said digital input signal.

16. Circuitry according to claim 15 when dependent on any one of claims 5 to 14, said circuitry further comprising:

a plurality of feedback paths for routing each of said plurality of amplified signal to at least one signal sampler;

at least one predistortion coefficient generating logic module comprising said predistortion coefficient generating logic configured to receive said sampled signals and said digital input signals corresponding to said sampled signals and to generate said predistortion coefficients from a comparison of said signals;

at least one of said data stores operable to store said generated coefficients for each of said plurality of predistortion logic modules along with an indicator of a statistical property of said corresponding input digital signal.

17. Circuitry according to claim 16, said circuitry comprising one of said signal sampler, one of said predistortion coefficient generating logic module and one of said data store, said circuitry further comprising:

switching circuitry for routing a selected one of said amplified signals to said one signal sampler; and

further switching circuitry for routing a corresponding selected one of said input digital signals to said one predistortion coefficient generating logic module.

18. Circuitry according to claim 15 or 16, said circuitry comprising:

a plurality of said signal samplers;

a plurality of predistortion coefficient generating logic modules each configured to receive a corresponding one of said sampled signals and said digital input signal corresponding to said sampled signal and to generate said predistortion coefficients from a comparison of said signals; and a plurality of said data stores for storing said generated coefficients for each of said plurality of predistortion logic modules along with an indicator of a statistical property of said input digital signals, said statistical property being received from said statistical property monitoring logic.

19. Circuitry according to any one of claims 15 to 18 , wherein said plurality of antenna comprise a massive MIMO antenna system .

20. A method for reducing distortions in an amplified signal to be radiated by an antenna, said method comprising:

monitoring at least one statistical property of a digital input signal;

selecting a predistortion function to apply to said input digital signal in dependence upon said monitored at least one statistical property of said digital input signal;

applying said selected predistortion function to said input digital signal.

21. A method according to claim 20 , wherein said step of selecting said predistorion function comprises selecting a set of digital predistortion coefficients for a predistortion model from a data store, each set being stored in said data store linked to a statistical property.

22. A method according to any one of claims 20 to 21, wherein said digital input signal comprises a wideband signal having a bandwidth greater than 100MHZ.

23. A method according any one of claims 20 to 22, wherein said step of monitorin is performed periodically and in response to determining a change in a monitored statistical property, said selecting step selecting an updated predistortion function to apply to said input signal.

24. A method according to any one of claims 20 to 23 , said method further comprising:

routing said amplified signal to a signal sampler on a feedback path ;

receiving said sampled signal and said digital input signal corresponding to said sampled signal and generating predistortion coefficients from a comparison of said signals; storing said generated coefficients along with an indicator of a statistical property of said input digital signal, said statistical property being received from said statistical property monitoring logic.

25. A method according to claim 24, said step of sampling further comprising converting an analogue signal to a digital signal.

26. A method according to claim 24 or 25, wherein said step of sampling comprises sampling said amplified signal at a sampling rate that is lower than a sampling rate of said input digital signal.

27. A method according to claim 26, wherein said sampling rate is lower than a Nyquist sampling rate for said amplified signal.

28. A method according to claim 26 or 27, wherein said comparison comprises performing convolution based peak detection to determine correlation levels between samples of said feedback signal and said digital input signals, a time offset between samples providing a highest peak of correlation being a time offset for shifting one set of signals to provide aligned pairs for comparison.

29. A method according to any one of claims 24 to 28 , wherein said comparison comprises a multi-rate least squares identification.

30. A method according to any one of claims 24 to 29, wherein said step of generating said predistortion coefficients is performed for input digital signals with different statistical properties.

31. A method according to any one of claims 24 to 30 , wherein said step of generation said predistortion coefficients comprises generating a lookup table indexed by statistical property for said coefficients during an initial training phase.

32. A method according to any one of claims 24 to 31, wherein said step of generating predistortion coefficients is performed periodically during operation.

33. A method according to any one of claims 24 to 32, said method further comprising monitoring said amplified signal and in response to detecting a portion of said signal outside a predetermined bandwidth updating said predistortion coefficients

34. A method according to any one of claims 20 to 33, wherein said method is for reducing distortions in a plurality of amplified signals to be radiated by a plurality of antennas, said method comprising:

monitoring at least one statistical property of a plurality of digital input signals; selecting a predistortion function to apply to a corresponding one of said plurality of digital input signals in dependence upon said monitored at least one statistical property of said digital input signal.

35. A computer program which when executed by a computer is operable to control said computer to perform a method according to any one of claims 20 to 34.

36. Means for reducing distortions in an amplified signal to be radiated by an antenna, said means comprising:

statistical property monitoring means for monitoring at least one statistical property of a digital input signal;

digital predistortion means for selecting a predistortion function to apply to said input digital signal in dependence upon said monitored at least one statistical property of said digital input signal.

37. Means according to claim 36, said means further comprising:

a signal sampling means for sampling said amplified signal;

a feedback path for routing said amplified signal to said signal sampling means; predistortion coefficient generating means for receiving said sampled signal and said digital input signal corresponding to said sampled signal and for generating predistortion coefficients from a comparison of said signals;

a data storage means for storing said generated coefficients along with an indicator of a statistical property of said input digital signal, said statistical property being received from said statistical property monitoring logic.