WO2010057974A1

WO2010057974A1 - Method and apparatus for locating channels and estimating channel baud rates

Info

Publication number: WO2010057974A1
Application number: PCT/EP2009/065550
Authority: WO
Inventors: Dirk Schmitt; Marten Kabutz; Alkis Ikonomu
Original assignee: Thomson Licensing
Priority date: 2008-11-21
Filing date: 2009-11-20
Publication date: 2010-05-27

Abstract

For determining a channel frequency estimate and/or a baud rate estimate, the system derives, by a frequency transform (113, 201), a spectral representation from data representing an incoming radio frequency spectrum; it is further adapted to smooth (202, 203) the spectral representation; to derive from the smoothed spectral representation, coarse band edge positions using a threshold detector (109); and to derive from the coarse band edge positions, refined band edge positions employing slope detection (110) and bandedge correlation (107) steps.

Description

METHOD AND APPARATUS FOR LOCATING CHANNELS AND ESTIMATING CHANNEL BAUD RATES

Technical field

The invention relates to Digital Communication, like Digital Video broadcast DVB, GSM, WLAN, or DSL. In particular it relates to estimating the bandwidth and frequency location of channels in a frequency multiplex multi-channel transmission system.

Background of the invention

Some digital communication systems employ frequency multiplexing also known as Frequency Division Multiple Access or FDMA to combine different payload signals also denoted as channels into an available spectrum. In this, if the baud rate of the individual payload signals differs, it is appropriate and feasible to assign to them distinct individual bandwidths . This approach is used in transmission systems like DVB-S, DVB- S2 and DVB-C. In the context of DVB, bandwidths of individual channels are known a priori for satellite and cable systems, but in the context of terrestrial broadcast under DVB-T, there are so called free-to-air applications where the frequency location and the bandwidth of the channels are not known. In contexts like these, a receiver needs to be able to scan the entire spectrum for new channels. Such scanning encompasses the following steps:

1. ) Finding the location of each used channel in the spectrum, especially its center frequency which is used in the tuner for down conversion. 2. ) Finding the bandwidth of each channel as found in step 1) .

Known methods for baud rate estimation employ a cross correlation of the incoming signal with its filtered version and evaluate the energy after the cross correlation. For this, a filter bank is used where the different correlation filters have bandwidths related to the possible transmission bandwidths This is summarized in Lee J. Y., Chung Y. M., Lee S. U., "ON A TIMING RECOVERY TECHNIQUE FOR A VARIABLE SYMBOL RATE SIGNAL", IEEE 1997 47^th Vehicular Technology Conference, pp. 1724-1728. This can be seen to have the disadvantage that the necessary filter bank represents a considerable implementation burden, regardless whether implemented in hardware or in software.

In WO 2007/050198 Al, an FFT and a spectrum analyzer are used to scan the spectrum in a blind scan manner. This method determines channel band edges by searching the FFT frequency bins for a IdB magnitude decrease. This can be seen to have the following disadvantages:

1.) For good detectability of band edges, a high resolution FFT using for example 2048 bins has to be used.

2.) Using a band edge criterion of a 1 dB decrease is not suitable for low SNR modes like in the DVB-S2 standard.

Summary of the invention

The present invention provides method and apparatus for determining channel estimates by scanning a spectrum using a spectrum analyzer and a post processing step for band edge detection . The scanning algorithm comprises several iterations, where every iteration works with a linear spectrum analyzing result.

For this, the spectrum has to be analyzed using an FFT or the so-called Goertzel algorithm with an averaging process (e.g. using Bartlett-Welch method) .

The post processing step depends on the expected noise floor, which could be measured using the estimated spectrum.

In case of a large distance between noise floor and signal energy, i.e. in high SNR cases, the current invention proposes a two step approach. The two steps are:

1.) Scanning the estimated spectrum using a threshold detector and a min hold logic for detecting the local minima that appear between the band edges of neighboring channels . 2. ) After at least one channel has been found in the spectrum, a slope detector is used to scan the spectrum for providing more precise information about the band edges.

The method is applied to the entire spectrum of interest. A microcontroller is used to shift the tuner frequency, so that the total spectrum could be divided into several overlapping or non-overlapping so-called scanning windows.

In case of a small distance between the noise floor and the signal energy, i.e. in low SNR cases, applying the two step approach as explained above would cause misdetections of band edges. For these cases the present invention proposes to use a three step approach, where a band edge correlation step is additionally applied before the spectrum analyzer. The band edge correlation indicates a peak in the spectrum if the band edges are centered around the zero frequency. For applying the band edge correlation step, the baudrate of the channel has to be known, or at least an assumption has to be made about it. But it gives a high noise rejectance which is helpful in very low SNR situations. Without a known baud rate, the three step approach can simply be used to scan for several widely used baud rates like 27 MBaud or 22.5 MBaud. Alternatively, a baud rate information obtained by applying the two step approach, although potentially less reliable, may be used as a baud rate estimate for applying the band edge correlation step.

Brief description of drawings

Figure 1 shows the overall block diagram of the channel estimates determining apparatus.

Figure 2 illustrates the main steps of the method according to the present invention.

Figure 3 shows signal magnitudes over frequency, which illustrate the first method for detection of the channel location and its bandwidth. Figure 4 shows a hardware implementation of the method for detection of the channel location and its bandwidth.

Figure 5 shows a modified moving average filter optionally inserted upstream of baud rate estimation. Figure 6 shows signal magnitudes over frequency, which illustrate the input and output behavior of the slope detector.

Figure 7 shows the bandedge correlator inserted between derotator and spectrum estimator.

Figure 8 (top) shows a scanning window with 4 populated channels, where only the center channell and channel2 are shown in their entirety. In Fig 8 (bottom) , peaks are only obtained if the filters are centered around the channell or channel2 band edges . Detailed description

Figure 1 shows the overall block diagram of the channel estimates determining apparatus. An incoming antenna signal 116 is multiplied in a first multiplier 102 with a first oscillator signal 114 having a frequency of ftuner+<delta>ftuner, and the product is lowpass filtered in a first lowpass filter 104. In parallel to this, the incoming antenna signal 116 is multiplied in a second multiplier 103 with a second oscillator signal 115 having the same frequency but a 90 degree shifted phase, and that product is lowpass filtered in a second lowpass filter 105, The numerically controlled oscillator, the two multipliers 102, 103, and the two lowpass filters 104, 105 together constitute what will also be called a derotator 101 in the following. The two output signals of the derotator 101 together constitute a quadrature signal, and are input to a quadrature analog to digital converter 106. Controlled by a switch Sl, the digital output of the analog to digital converter 106 is either forwarded directly to an FFT processor 113, or is looped through a band edge correlator 107 before being input to the FFT processor 113. The complex valued output bins of the FFT processor 113 are input to a magnitude estimator 112, and the real valued output of the magnitude estimator 112 is made available for a min holder 109, for a slope detector 110, and for a peak detector 111. The min holder 109, the slope detector 110, and the peak detector 111 all are controlled by a microcontroller 108. The micro-controller 108 also controls the frequency ftuner+<delta>ftuner of the numerically controlled oscillator .

Due to a combination between parts implemented in hardware and other parts implemented in software, the apparatus according to the invention amounts to a small hardware impact for baud rate estimation. Compared to the known solutions mentioned above the methods and apparatuses according to the invention have the following advantages:

1.) The high HW impact of a filter bank as needed in the correlation methods is avoided.

2.) The slope detection step and the band edge correlation step together provide a detection quality which outperforms the method of WO 2007/050198 Al.

As mentioned above the invention locates channels and estimates channel baud rate using a two step approach. Also a three step approach is proposed using the band edge correlation. The latter can be used if the baud rate is known but the channel location is not.

The two step approach comprises two stages. The first stage consists of an FFT 113 which estimates the spectrum of short subsequences extracted from the incoming signal. In order to smooth the estimated spectrum, subsequences are extracted and multiplied with a window function using the known Bartlett Welch method, and the magnitudes of FFT bins are averaged over several such windows. For this, the magnitude mag of an FFT bin x is estimated in the magnitude estimator 112 as

mαg = max(reαl(x),imαg(x)) + 0.5 *mm(reαl(x),imαg(x)) where real (x) is the real part of the FFT bin x and imag(x) is the imaginary part of the FFT bin x. The smoothing window length can be adjusted by a software register under control of the micro-controller 108. Smoothing the spectrum by averaging several FFT results saves HW impact and gives the ability to trade off calculation time against accuracy of the magnitude estimates. The results of the estimated spectrum is then stored in a memory for further processing.

Figure 2 illustrates the main steps of the method according to the present invention. In step 201, an FFT is performed, and in step 202, the result of the FFT is averaged with results of previously performed FFTs. Step 203 constitutes the loop control, where the loopback criterion whether the last window has been reached, is tested. If the last window has not been reached, the program loops back to step 201, otherwise it proceeds to a subsequent step 204. In the step 204, the spectrum obtained by averaging a number of FFT output bins is scanned using the min holder 109 or using the slope detector 110. In a subsequent step 205, the output of the scanning step 204 is searched for peaks using the peak detector 111. In a subsequent step 206, the peak locations identified by the peak detector 111 are stored.

In the second stage, two different algorithms can be used for detection of the channel location and its bandwidth. The two algorithms can be implemented in software or in hardware.

Figure 3 shows signal magnitudes A over frequency f, which illustrate the first method for detection of the channel location and its bandwidth, which uses a brute-force approach. In Figure 3, a continuous line 301 indicates a hypothetical magnitude of FFT bins over frequency f as provided by the magnitude estimator 112. The min holder 109 searches from left to right for minima in the magnitude spectrum 301 and writes lower bits of the detected min values to a shift register. A first minimum 304 and a second minimum 305 have been found. Scanning from left to right, i.e. from low frequencies to high frequencies is assumed, but the method could as well be implemented with scanning from right to left or with scanning in both directions alternatingly . The detected minima are stored in a memory and will be evaluated in the further processing.

Figure 4 shows a hardware implementation of the method for detection of the channel location and its bandwidth. The hardware consists of a first comparator 401 which compares a current magnitude 2, which is the estimated magnitude of a currently scanned spectrum bin, with a first threshold value 1. If the current magnitude 2 is less than the threshold 1, a first shift register 402 of a chain of shift registers 402, 403, 404 gets enabled to store the current position 3 and the value of the current magnitude 2.

On the other hand, if the current magnitude 2 is bigger than a second threshold value 4, shift register content is only forwarded without inscribing a new value into the first shift register 402. Forwarding means that the content of the first shift register 402 is forwarded to a second shift register 403, the content of the second shift register 403 is forwarded to a third shift register 404, and so on.

Conceptually, the shift register is shifted if a new band edge of an adjacent channel is detected. The two threshold comparisons 401, 405 constitute a kind of hysteresis to avoid false detections which would otherwise result from noise variations .

The following processing step uses the information of the band edge locations to decide how many channels are populated in the spectrum window, and to coarsely estimate their band edges. Figure 5 shows a modified moving average filter optionally inserted upstream of baud rate estimation to smooth the spectrum. The filter runs through the estimated spectrum window, The modified moving average filter receives an input sequence 509, which is processed in a finite impulse response or FIR section, and subsequently in an infinite impulse response or HR section, to yield an output sequence 506. The FIR section consists of a chain of feed forward registers 501, ... , 503; and the HR section consists of a single feedback register 505. The FIR section and the HR section are connected by an adder 504, which receives the output 510 of the first feed forward register 501, subtracts thereof the output 507 of the last feed forward register 503, and adds thereto the output 508 of the feedback register 505. By receiving the output 510 of the first feed forward register 501 and subtracting thereof the output 507 of the last feed forward register 503, the adder 504 acts as an FIR filter. Then the output 508 of the feedback register 505 is added thereto, which embodies a simple HR filter.

Figure 6 shows signal magnitudes A over frequency f, which illustrate the input and output behavior of the subsequent slope detector 110. Slope detection constitutes the second step which can be used as the fine estimation step of band edges. Proceeding through all frequency bins, the slope detector 110 measures the slope between pairs of magnitude values from the spectrum, whose frequencies are apart by an arbitrarily chosen constant. The upper part of Fig. 6 shows a hypothetical magnitude spectrum as the input of the slope detector 110, and several superimposed double arrows that symbolize magnitude pairs between which the slope is being measured. The bottom part of Fig. 6 shows the frequency characteristic of the slope that is correspondingly being measured as the output of the slope detector 110. When scanning from left to right, a positive peak is obtained if the slope detector passes a low frequency band edge, and a negative peak is obtained at a high frequency band edge. If the bandwidth of the channel is known, slope detection can be improved by averaging between the slope characteristic at the lower edge and the inverted slope characteristic at the upper edge, and then constitutes an improved channel location estimator.

The slope detector result is stored in a max hold and min hold register, to be used for subsequent estimation of the channel carrier and the bandwidth of the detected channel. If detection of multiple channels is unwanted, the scanning range in the current spectrum can be limited and the scanning window can be divided into several sub scanning windows if more then one channel is assumed to be in the current scanning window.

Figure 7 shows the bandedge correlator 107 inserted between derotator and spectrum estimator. This configuration is suited for low SNR modes where the noise floor is just slightly below the signal spectrum. For this method the baud rate - and hence the bandwidth - has to be known with high accuracy. A signal 703 coming from a numerically controlled oscillator and signal derotator 701 corresponding to 101 in Fig. 1 is fed in parallel to two complex filters 702, 708. The first complex filter 702 is tuned centered around the estimated band edge and the second complex filter 708 is just a delay element, which corresponds to an all-pass filter. From the output xl (t) of the first complex filter 702 and the output x2 (t) of the second complex filter 708, a complex multiplier 704 calculates a product y(t) according to

y(t) = x_l(t)*x₂*(t), where x2* denotes the conjugate complex of the output x2 of the delay element 708. Because taking the complex conjugate corresponds to reversing the spectrum, the output y(t) yields a peak 803 if filter Fl 702 is centered around the lower band edge and filter F2 708 is centered around the upper band edge. Further, the output from the multiplier 704 is subjected to dc offset removal 706. The first complex filter 702, the second complex filter 708, the multiplier 704, and the dc offset removal 705 together constitute the bandedge correlator 107.

As Figure 1 had shown, the output of the bandedge correlator 107 is fed to the spectrum analyzer denoted as FFT processor 113 in Fig. 1 and as spectrum estimation 706 in Fig. 7.

In the further scanning process the bandedge correlator 107 processes the scanning window by incrementing the frequency offset of the digital NCO in front of the band edge correlator. Every time the band edge filters are centered around two band edges, the output of the bandedge correlator has a peak 803, which will be detected by a peak detector. First the scanning window is fed to an up/down conversion logic 701 comprising an NCO and a digital derotator which is able to shift the frequencies in the scanning window up and down. This is needed to shift the band edges to the right location in the spectrum where the following filters are centered around. If needed, the up down converters can be avoided by using two complex filters. The up/down converted results are then fed to the two band edge filters 702, 708. The output of the filters is then multiplied in a conjugate complex manner to each other and the result is fed to a dc offset compensation logic 705. This logic can be a simple averager which averages the incoming signal for a short time and subtracts the result from the input values to get a dc free output. The output of the dc offset removal 705 is then fed to the spectrum analyzer 706 and to a peak detector 707

(e.g. using a threshold detector). The peek detector 707 evaluates the output of the spectrum analyzer 706 to catch the peak if the filters Fl 702 and F2 708 are centered around two band edges which belong together. As mentioned above the scan process is using the NCO to move the frequencies regarding to the two filter locations. Also the two filters could be moved by setting their coefficients.

Figure 8 (top) shows the scanning window with 4 channels, where only channels indicated as "chl" and "ch2" are fully depicted. In the bottom part of Fig. 8, arrows symbolize those locations where the filters of the bandedge correlator are centered around the band edges of channel "chl" or "ch2". The arrows are grouped into a first group 806 near the band edge between channel chO and channel chl, and a second group 807 near the band edge between channel chl and channel ch2.

The general idea of the invention is applicable to any baud rate estimation in any digital communication. The band edge correlation algorithm can be used for raised cosine transmission, for root raised cosine transmission with a roll off, or if there is a gap between adjacent channels.

Claims

1. A method for determining, as a channel estimate, at least one of a channel frequency estimate and a baud rate estimate from data representing an incoming radio frequency spectrum, the method comprising

- deriving, by a frequency transform (113, 201), a spectral representation from the data;

- smoothing (202, 203) the spectral representation; - deriving, from the smoothed spectral representation, coarse band edge positions using a threshold detector (109),

- deriving, from the coarse band edge positions, refined band edge positions as the channel estimate employing slope detection (110) and bandedge correlation (107) steps.

2. A method according to claim 1, wherein the bandedge correlation step comprises filtering a signal (703) with a first complex filter (702), filtering the signal (703) with a second complex filter (708) which is an allpass, and calculating the product (704) of the output of the first complex filter (702) and the conjugated complex output of the second complex filter (798).

3. A method according to one of the previous claims, wherein deriving the spectral representation is performed using an FFT or a Goertzel algorithm.

4. A method according to one of the previous claims, wherein smoothing the spectral representation is performed using the Bartlett-Welch algorithm.

5. An apparatus for determining, as a channel estimate, at least one of a channel frequency estimate and a baud rate estimate from data representing an incoming radio frequency spectrum, the apparatus comprising

- frequency transform means (113), which receive the data at their input and derive a spectral representation of the data at their output,

- smoothing means (202), which receive the spectral representation at their input and derive a smoothed spectral representation at their output; and characterized by - nonlinear postprocessing means (109,110,111), which receive the smoothed spectral representation at their input and derive the channel estimate, and

- bandedge correlation means (107) which receive the digitalized data, are equipped and configured to apply a first complex filter (702) and a second complex filter (798) to the data and to calculate the product between the output of the first complex filter (702) and the conjugate complex of the output of the second complex filter (708) .