HK1124973B - Equalizer for am in-band on-channel radio receivers - Google Patents

Description

Equalizer for AM in-band on-channel radio receiver

Technical Field

The present invention relates to radio broadcasting, and more particularly, to a method and apparatus for equalizing a signal in a receiver for an in-band on-channel digital broadcasting system.

Background

An AM compatible in-band on-channel (IBOC) digital broadcast system simultaneously broadcasts analog and digital signals within a standard AM broadcast channel. An AM IBOC system is described in U.S. patent No.5,588,022. The broadcast signal includes an amplitude modulated radio frequency signal having a first frequency spectrum. The amplitude modulated radio frequency signal includes a first carrier modulated by the analog program signal. The signal also includes a plurality of digitally modulated carrier signals within a bandwidth containing the first frequency spectrum. Each digitally modulated carrier signal is modulated by a digital signal. The first set of digitally modulated carrier signals are within a first frequency spectrum and are quadrature modulated with the first carrier signal. The second and third sets of digitally modulated carrier signals are outside the first frequency spectrum and are in-phase and quadrature modulated with the first carrier signal. The subcarriers are subdivided into first, second and third level partitions. Some of the subcarriers are complementary subcarriers.

Equalization of the received multicarrier signal is required when there is a dynamic channel response variation. Without such equalization, distorted signals would be detected, rendering the digital broadcast signal information unrecoverable. The equalizer enhances the recoverability of the digital audio broadcasting signal information. In U.S. patent nos. 5,559,830; 6,292,511, respectively; 6,295,317, respectively; and 6,480,536, an equalizer in a receiver for receiving AM in-band on-channel signals is disclosed.

The use of complementary subcarriers for mixing the second and third level partitions in an AM compatible digital audio broadcast signal results in an orthogonal relationship with the analog host signal. Existing equalization implementations for second level partitioning require knowledge of whether to limit the analog host bandwidth to ± 5 kHz. If the analog host bandwidth is limited to 5kHz, the second level partitions are equalized independently to better account for adjacent channel interference. Otherwise, the second partitions are first complementarily combined to cancel the analog signal in that partition.

Thus, there is a need for an equalization technique that does not require analog bandwidth information.

Disclosure of Invention

The present invention provides a method for equalizing OFDM symbol vectors received on an AM in-band on-channel radio signal comprising a main carrier and first and second BPSK modulated subcarriers. The method comprises the following steps: calculating a BPSK magnitude signal; filtering the BPSK magnitude signal; filtering the complex samples received on the main carrier; calculating a plurality of flat fading equalization coefficients using the filtered BPSK magnitude signal and the filtered complex samples received on the main carrier; and multiplying the OFDM symbol vector by the flat fading equalization coefficient.

In another aspect, the present invention provides a method of equalizing a vector of OFDM symbols received on an AM in-band on-channel radio signal, the method comprising the steps of: setting a plurality of training symbols in a training symbol vector; median filtering the training symbol vectors to generate median estimates of the training symbol vectors; smoothing the median estimate of the training symbol vector in time and frequency; calculating a plurality of equalization coefficients using the smoothed median; and multiplying the OFDM symbol vector by the equalization coefficient.

In yet another aspect, the present invention provides a method of estimating a variance of training symbol information received on an AM in-band on-channel radio signal, the method comprising the steps of: setting a plurality of training symbols in a training symbol vector; calculating the logarithm of the local estimation variance of the training symbol vector; smoothing the logarithm of the variance of the training symbol vector in time and frequency; a plurality of channel state information values are calculated using the smoothed estimate of the logarithm of variance.

The invention also relates to a method of equalizing OFDM symbol vectors received on AM in-band on-channel radio signals, said method comprising the steps of: complementary combination is carried out on the second-stage subarea and/or the third-stage subarea in the AM in-band on-channel radio signals; calculating a plurality of flat fading equalization coefficients, and multiplying the OFDM symbol vector by the flat fading equalization coefficients to generate a flat fading equalized OFDM symbol vector; and calculating a plurality of partition equalization coefficients and multiplying the flat fading equalized OFDM symbol vector by the partition equalization coefficients to generate an output OFDM symbol vector.

There is also provided a receiver comprising an equaliser operating in accordance with the above method.

Drawings

Fig. 1 shows a spectral diagram of an AM hybrid IBOC signal.

Fig. 2 shows a spectral diagram of an AM all-digital IBOC signal.

FIG. 3 shows a functional block diagram of an AM IBOC receiver.

Fig. 4 shows a block diagram of a modem for an AM IBOC receiver.

Fig. 5 shows a block diagram of a flat fading equalizer constructed in accordance with the present invention.

Fig. 6 is a block diagram of a partitioned equalizer constructed in accordance with the present invention.

Detailed Description

Referring to the drawings, fig. 1 shows a spectral diagram of an AM hybrid IBOC signal. The AM hybrid IBOC waveform 10 includes a conventional AM analog signal 12 (band limited to about ± 5kHz), and an approximately 30kHz wide Digital Audio Broadcast (DAB) signal 14 transmitted below the AM signal. The spectrum is contained within a channel 16 having a bandwidth of about 30 kHz. The channel is divided into a center frequency band 18, an upper frequency band 20, and a lower frequency band 22. The central frequency bandwidth is about 10kHz and includes a frequency f from the center of the channel_oA frequency of about ± 5 kHz. The upper sideband extends from about +5kHz from the center frequency to about +15kHz from the center frequency. The lower sideband extends from about-5 kHz from the center frequency to about-15 kHz from the center frequency.

In one embodiment of the present invention, the AM hybrid IBOC DAB signal format includes an analog modulated carrier signal 24 and 162 OFDM subcarrier locations spaced approximately 181.7Hz apart that span the center band and the upper and lower sidebands. Encoded digital information representing an audio or data signal (program content) is transmitted on the subcarriers. The symbol rate is less than the subcarrier spacing due to the guard time between symbols.

As shown in FIG. 1, the upper sideband is divided into a first level partition 26 and a second level partition 28, and the lower sideband is divided into a first level partition 30 and a second level partition 32. The digital signals are transmitted under the host analog signal in the first and second level partitions on either side of the host analog signal, and in the third level partition 34. The tertiary partition 34 may be considered to include multiple sets of subcarriers, as shown in fig. 1, labeled 36, 38, 40, and 42. Subcarriers within a tertiary partition located closer to the center of the channel are referred to as inner subcarriers, and subcarriers within a tertiary partition located farther from the center of the channel are referred to as outer subcarriers. In this example, the power levels of the inner subcarriers in groups 38 and 40 are shown to decrease linearly with increasing frequency spacing from the center frequency. The remaining sets of subcarriers 36 and 42 in the tertiary sideband have substantially constant power levels.

Fig. 1 also shows two reference subcarriers 44 and 46 for system control, which are located in the first subcarrier position next to the analog modulated carrier and whose power levels are fixed at values different from the other sidebands.

At frequency f_oThe center carrier 24 is not QAM modulated but carries a main analog amplitude modulated carrier. The synchronization and control subcarriers 44 and 46 are modulated orthogonally to the carrier. The remaining subcarriers of the third level segment at locations denoted 2 through 26 and-2 through-26 on either side of the AM carrier are modulated with QPSK. Typical subcarrier locations are identified by subcarrier indices as shown in fig. 1. The subcarriers at positions 2 through 26 and-2 through-26 on either side of the center frequency are referred to as tertiary subcarriers and are transmitted in complementary pairs so that the resulting DAB signal is modulated in quadrature with the analog modulated AM signal. The use of complementary subcarrier pairs in AM IBOC DAB systems is shown in U.S. patent No.5,859,876. In addition, synchronization and control subcarriers 44 and 46 are also modulated into complementary pairs.

The Double Sideband (DSB) analog AM signal occupies a bandwidth of the ± 5kHz region. The lower and upper tertiary partitions occupy sub-bands from about 0 to about-5 kHz and from about 0 to about +5kHz regions, respectively. These tertiary partitions are negative complex conjugates of each other and have complementary properties. This complementary property maintains the quadrature relationship between the analog and digital third stage signals so that they can be separated in a receiver while existing conventional receivers can still receive the analog AM signal. The third stage partitions must be complementarily combined to extract the digital signal while eliminating analog crosstalk noise. The secondary partitions also have complementary properties so that they can be processed at the receiver, either independently or after complementary combining, depending on the interference conditions and audio bandwidth. The first level partitions are sent independently.

Fig. 2 shows a spectral diagram of an all-digital IBOC signal 50. The power of the center band 52 subcarriers is increased relative to the hybrid format shown in fig. 1. Likewise, the two subcarriers 54 and 56 at positions-1 and +1 transmit timing information using binary phase shift keying. The core upper sideband 58 includes carriers at positions 2 through 26 and the core lower sideband 60 includes subcarriers at positions-2 through-26. Sidebands 58 and 60 form the first level partitions. Two sets of additional enhanced subcarriers 62 and 64 occupy positions 27 through 54 and-54 through-27, respectively. Group 62 forms the second level partition and group 64 forms the third level partition. The all-digital format shown in fig. 2 is very similar to the hybrid format except that the AM signal is replaced by a delayed digitally encoded tuned and backup version of the program content. In the hybrid and all-digital formats, the center frequency band occupies approximately the same spectral location. In the all-digital format, there are two options for sending the primary version of the program content in conjunction with the tuning and backup versions. Designing an all-digital system to be constrained to a channel center frequency f_oWithin + -10kHz, wherein, at f_oMain audio information is transmitted within a range of ± 5kHz, while less important audio information is transmitted at a low power level within a range of ± 10kHz out from the channel mask side. This format allows for a large reduction of the signal while increasing the coverage area. All-digital systems carry digital time diversity tuning and backup channels within a 5kHz protection area (assuming digital audio compression is capable of delivering main and audio backup signals within the 5kHz protection). All-digital systemIs based on an AM IBOC hybrid system.

The all-digital IBOC signal includes a pair of first stage partitions in the +5kHz region, a second stage partition in the-5 kHz to-10 kHz region, and a third stage partition in the +5kHz to +10kHz region. The all-digital signal has no analog component and all partitions are sent independently (i.e., the partitions are not complementary).

Fig. 3 shows a functional block diagram of an IBOC receiver 84 constructed in accordance with the present invention. The IBOC signal is received on antenna 86. The band-pass preselection filter 88 passes the frequency band of interest, including at frequency f_cBut removed at f_c-2f_ifImage signal (for lower sidelobe injection local oscillator). The low noise amplifier 90 amplifies the signal. The amplified signal is mixed in mixer 92 with a local oscillator signal f provided by a tunable local oscillator 96 on line 94_loMixing is performed. This generates a sum (f) on line 98_c+f_lo) Sum and difference (f)_c-f_lo) A signal. IF filter 100 applies an IF signal f_ifAnd attenuates frequencies outside the bandwidth of the modulation signal of interest. The analog-to-digital converter 102 uses the clock signal f_sOperating at a rate f_sDigital samples are generated on line 104. Digital down converter 106 frequency shifts, filters, and samples the signal to generate low sample rate in-phase and quadrature signals on lines 108 and 110. A digital signal processor based demodulator 112 then provides additional signal processing to generate an output signal on line 114 for an output device 116.

The receiver shown in fig. 3 includes a modem constructed in accordance with the present invention. FIG. 4 shows AM HD Radio^TMA functional block diagram of the modem 130 showing the functional location of the carrier tracking of the present invention. The input signal on line 132 from the digital down converter undergoes carrier tracking and automatic gain control as shown in block 134. The resulting signal on line 136 is subjected to a symbol tracking algorithm 138, which generates a BPSK signal on lines 140 and 142, a symbol vector on line 144 (in the time domain),and an analog modulated carrier on line 146. BPSK processing as shown in block 148 generates block/frame synchronization and mode control information 150 which is used by the functions shown in the other blocks. OFDM demodulator 152 demodulates the time domain symbol vectors to generate frequency domain symbol vectors on line 154.

The equalizer 156 processes the frequency domain symbol vectors in conjunction with the BPSK and carrier signals to generate an equalized signal on line 158 and channel state information on line 160. The signals are processed to generate branch metrics 162, deinterleaved in a deinterleaver 164, and mapped in a deframer 166 to generate soft decision bits on line 168. The Viterbi decoder 170 processes the soft decision bits to generate decoded program data units on line 172.

For clarity of illustration, the OFDM vectors are distinguished into time and frequency domain vectors, each representing the same information. The modem processes these OFDM vectors in the following order (see fig. 4): carrier tracking, symbol tracking, OFDM demodulation & BPSK processing, and then equalization. The input to the modem comprises a time domain vector, or simply a sequence of time samples; the carrier tracking operates in the time domain. Symbol tracking operates on time domain samples and outputs a time domain OFDM vector (symbol synchronized), and furthermore, the middle 3 FFT bins (0, ± 1) representing the main carrier and BPSK subcarriers in the frequency domain are computed. The main carrier and BPSK subcarriers are used for equalization for reception from symbol tracking, but they may also be received from the OFDM demodulation function (windowed FFT) with the same redundant 3 intermediate FFT bins. The equalizer always operates on frequency domain OFDM vectors.

The present invention relates to a method and apparatus for equalizing a hybrid or all-digital AM IBOC signal. The equalizer consists of two cascaded elements, a flat fading equalizer followed by a partitioned equalizer, and uses noise variance estimation in generating Channel State Information (CSI). Flat fading compensation is applied in the same manner to both the hybrid and all-digital signals. The partition equalizer operates on each partition of the received signal. In one example, each partition consists of a group of 25 OFDM subcarriers, each partition spanning approximately 5 kHz. The partitions of the all-digital IBOC signal include a pair of first, second, and third stage partitions, and are independently equalized. However, the second and third level partitioning of the mixed signal includes additional processing and combining techniques described below. Several other single subcarriers (singlesubsarriers) are also sent between partitions and use simpler equalization techniques than those described herein.

Flat fading compensation (equalizer) is described below. The flat fading compensation includes phase compensation using the phase of the main carrier, and amplitude equalization using the imaginary component of the BPSK signal. This flat fading compensation should be applied to all OFDM subcarriers.

Consider a single digital QAM (complex) symbol (Q (n, 1) ═ x + j · y), and an analog signal component (a (n, 1) ═ u + j · v) of the AM IBOC signal. The symbol is at a subcarrier frequency f in the nth OFDM symbol_cOne symbol of the transmitted QAM symbol set. QAM symbols are transmitted using complementary subcarrier pairs to avoid AM crosstalk.

s(t)＝[a(n，1)+Q(n，1)]·w(t)·e^j·2·π·f _c ^·t+[a(n，-1)+Q(n，-1)]·w(t)·e^-j·2·π·f _o ^·t

Wherein Q (n, -1) ═ Q^*(n, 1) and a (n, -1) ═ a^*(n，1)。

The receiver demodulates the signal (which has been added to the analog modulation component and corrupted by noise and phase errors) to generate symbol estimates for the subcarrier pairs, as follows:

D(n，1)＝∫(s(t)+n(t))·w(t)-e^-j·2·π·f _c ^·t+j·φ·dt＝Q(n，1)·e^j·φ+a(n，1)·e^j·φ+n₁

D(n，-1)＝∫(s(t)+n(t))·w(t)·e^j·2·x·f _c ^·t+j·φ·dt＝Q(n，-1)·e^j·φ+a(n，-1)·e^j·φ+n_-1。

to show the effect of the complementary combination, the analog component can be extracted by adding the two components. The analog signal can be regenerated using the real part of the result or, more commonly, its amplitude is calculated.

When phi and noise are sufficiently small.

Extracting the digital symbol as

When phi and noise are sufficiently small.

The BPSK sequence is transmitted on a first pair of OFDM subcarriers on either side of the main carrier. At a gain G relative to the main carrier at level 1_BPSKThese BPSK subcarriers are transmitted. Thus, each BPSK symbol may be recovered and adjusted by the following expression:

when phi and noise are sufficiently small.

Then, what is of interest is an estimate of the absolute value of the BPSK bits (real scalar) which are used for subsequent scaling of the signal. For this particular BPSK symbol, where Q (n, 1) ═ x + j · y, where x is arbitrarily chosen to be 0, and the information bits are applied in the imaginary dimension. To extract scalar information r (n) from b (n), its magnitude may be calculated, or the absolute value of imaginary component y may be extracted from b (n).

R (n) ═ b (n) |, or r (n) ═ abs [ Im { b (n) }.

When the phase error is small, the amplitude estimation is generally less accurate than the imaginary calculated absolute value. The computation on the amplitude is more complex, so we choose to avoid amplitude computation, rather than favor imaginary component computation. The estimate of the channel amplitude R (n) may be calculated from B (n), or more directly from D (n, 1) and D^＊(n-1).

Note that r (n) is a real-valued scalar.

A functional block diagram of the flat fading equalizer 180 is given in fig. 5. Input d (n) from the OFDM demodulator is provided on line 182. In this embodiment, the input is a 256 sample vector for each symbol n. As shown, r (n) is calculated in block 184 and passed to a median filter 186 to generate a first filtered signal on line 188. The first filtered signal is also filtered via a finite impulse response filter 190 to generate a second filtered signal on line 192.

In this embodiment, the filtering for the BPSK magnitude signal r (n) includes a 7-tap median filter in cascade with a 7-tap FIR filter. The median filter can be implemented by setting the samples of r (n) in a 7-element circular buffer and then calculating the median of the 7 samples. The median filter has a delay of 3 samples. The 7-tap FIR filter has a delay of 3 samples and is implemented by using the following 7 coefficients:

the total delay of the median and FIR filters is 6 samples. The filtered channel amplitude can be expressed as:

where the median is calculated over 7 samples.

In addition, the phase of the main carrier is also corrected to a flat fading component. However, the phase should be filtered independently of the preceding BPSK magnitude. This is due to the increased phase noise on the carrier samples approaching pinch-off (pinchoff) at the negative analog modulation peaks. The same FIR filtering 194 defined for BPSK amplitudes can be used for the main carrier phase but median filtering should not be used but instead an equivalent delay 196 to match the delay of the amplitude component. The primary carrier samples c (n) may be calculated independently on each OFDM symbol, or values calculated in OFDM demodulation may be used. The filtering for the primary carrier component is as follows:

the flat fading equalizer weights are the inverse of the filtered channel amplitude (divided by the zero guard epsilon) while applying the conjugate of the main carrier phase, after a suitable delay,

as shown in block 198.

The original input is delayed, as shown in block 200, and multiplied by W_ffAs shown in multiplier 202 to generate an output 256-sample vector on line 204 for each new symbol n-6 after flat fading equalization.

For calculating flat fading equalization coefficients W for each new OFDM symbol_ffThe algorithm of (a) is summarized as follows:

'Flat fading equalization algorithm'

Calculating the nominal value of amplitude R (n) of BPSK signal to be 1

Filtering 7 sampled medians with delay of 6 symbols

Sample (complex) filtering the main carrier with delay of 3 symbols

Flat fading coefficients for multiplication with OFDM symbol subcarriers are calculated, delay being 6 symbols.

The filtering for the BPSK magnitude signal r (n) includes a 7-tap median filter in cascade with a 7-tap FIR filter.

The above-described flat fading equalization is followed by partition equalization. Table 1 shows the positions (indices) of the interleaved symbols, including the training symbols "T" within each partition block. Each column represents a partition.

TABLE 1

Second, an algorithm is used to calculate the equalizer coefficients and associated noise variance that are estimated for each of the 25 elements (columns of subcarriers) of each OFDM symbol within a partition (e.g., the first level partition above). The equalizer begins processing OFDM symbols as they are received. For each OFDM symbol containing 25 columns (per partition), all partitions of the full digital mode and the first level partition of the hybrid mode are processed independently. The hybrid second stage partitions are processed independently, allowing the maximum metric to be selected after complementary combining, depending on whether the analog audio bandwidth is limited to 5 kHz. Only after complementary combining, the hybrid third level partitions are processed.

Each column of a partition contains 1 or 2 training symbols (complex numbers), depending on which of the 16 rows is processed. The training symbol positions are repeated every 16 OFDM symbols (rows). The position of the training symbols is typically computed as a function of the particular row of OFDM symbols (modulo 16). Next, the latest training symbols are collected along a 25-column vector TS, and only the column of TS (col) corresponding to the latest column of OFDM symbols containing training symbols is updated. After updating the neighbor set with the latest training symbols, the median and variance of the neighbor set are calculated. Next, the square and median values are filtered using a two-dimensional recursive filter technique. The equalizer coefficients are calculated from the filtered median and equalization is applied to all corresponding columns of the previous OFDM symbol and the updated noise variance (and inverse) is used for subsequent symbol processing. Details of this process will be described below and given in fig. 6.

Fig. 6 shows a functional block diagram of an equalizer that can be used for each 25-column partition. The OFDM symbols OFDM (r, col) are input on line 210. Training symbols are collected as indicated in block 212. The median and variance are calculated, as shown in block 214, to generate a median and variance signal on line 216. These signals are filtered and equalized in block 218 to generate an equalized variance signal on line 220 (for subsequent Channel State Information (CSI) estimation) and equalization coefficients on line 222. After a delay as shown at block 224, equalization coefficients are applied to the input signal as shown at block 226 to generate an output signal on line 228.

To compute the median and variance from the Training Symbols (TS), first, two 1-row 25-column matrices, labeled TS and MED, are created for storing the training symbols and the median computation, respectively. The column indices (col ═ 0 to 24) are equal to the corresponding columns of training symbols since they are received for each OFDM symbol. Next, the element is initialized to 0.

Then, the next OFDM symbol row r (modulo 16) corresponding to a particular row (r) of the interleaved block is received. The training symbol position or column is identified and for that row r, the training symbols are placed into the corresponding ts (col). The training symbols in row r may be updated using the following algorithm.

The partitioned equalizer performs several steps.

Step 1: the training symbols are aggregated, compressed (collapse) and updated into a vector TS (representing the timely training symbol information) for use in subsequent equalization processing.

"Algorithm for updating TS of Row r"

col ═ mod (3 · r +1, 16); "identifying which column has a new TS"

TS(col)＝OFDM(r，col)

ifcol < 9 then TS (col +16) ═ OFDM (r, col + 16); "if it is the second TS in the row".

Step 2: two 25-column vectors, labeled MED and logVAR, are created for storing the calculated median and the logarithm of the variance values for equalization and CSI. Median filtering of the local (time & frequency) TS samples is used to generate a median estimate for the TS. The outputs Med and logVAR are local estimates of these parameters (not yet time or frequency (over subcarriers) smoothed).

The column indices are equal to the corresponding columns of training symbols since they are received for each OFDM symbol. The element is then initialized to 0.

The median and variance are calculated for the ts (col)6 rows after updating the specific ts (col). This delay ensures that its neighboring training symbols are also updated for subsequent calculations. The 1 or 2TS (col) value is updated for each new row r. The 9-sample median and variance are calculated for columns 4 to 20 using ± 4 values on either side of the training symbol. For example, the median calculation for column 4 uses training symbols TS (0) to TS (8). Columns 0 to 3 and 21 to 24 are special cases since there are less than 9 samples available at the end to calculate the median and variance values. The end (extreme) missing values are replaced by copying values by folding near the end, if necessary. For example, in calculating the median value for column 3, TS (0) through TS (7) are used, replacing the missing TS (-1) column with TS (0) to provide 9 values for the median calculation. The calculated median and variance values are set in med (col) and logvar (col). The following method (pseudo-code) may be used to identify the appropriate column at this row r for updating and to aggregate the appropriate TS samples for the 9-sample median and log variance calculations:

"Algorithm for updating MED and logVAR vectors, delay 6 symbols"

col ═ mod (3 · r +15, 16); "first TS column recognizing r-6"

FOR m is 0 to 8; "aggregation of 9 adjacent TSs into buffer for MED & logVAR calculation"

colm＝col+m-4

TSmedbuff(m)＝TS(TScolindx)

Med (col) ═ mean (tsmedbuff); "Complex median, separating real and imaginary parts" "" computing the base 2 logarithm for the VAR samples (vectors) "

ifcol is less than 9 then; "update the second TS in this row, if any"

col2＝col+16

FOR m＝0 to 8

colm＝col2+m-4

TSmedbuff(m)＝TS(TScolindx)

MED (col2) ═ mean (tsmedbuff); "Complex median, separating real and imaginary parts"

end if

To calculate the equalizer coefficients and Channel State Information (CSI), the next step is to smooth (filter) the median and variance values in time and frequency (columns). The logarithm of the variance is used to smooth squared noise samples with potentially large dynamic range over the subcarriers.

Two 25-column vectors, labeled MED1 and logVAR1, were created for storing the recursive temporal filtering median and logarithmic variance values, respectively. The column indices are equal to the corresponding columns of training symbols since they are received for each OFDM symbol. The element is then initialized to 0.

Two 25-column vectors are created, labeled MED2 and logVAR2, for storing column or frequency filtered median and log variance values, respectively. The column indices are equal to the corresponding columns of training symbols since they are received for each OFDM symbol. The element is then initialized to 0.

The equalizer value EQ is calculated from the MED 2. The EQ value is typically the complex reciprocal of the MED2 value, but with division protection. The variance value logVAR2 is used to calculate the VAREQ for the subsequent CSI and branch metrics after equalizer gain adjustment.

And step 3: next, the MED and logVAR values are smoothed over time and frequency (subcarriers). Temporal smoothing through an IIR filter results in the generation of MED1 and logVAR 1. Frequency smoothing using one of the quadratic fit functions (quadratic fit fusion) resulted in the production of MED2 and logVAR 2. See the first part of the algorithm described below.

And 4, step 4: the equalizer value EQ is calculated by the MED 2. The EQ value is typically the complex reciprocal of the MED2 value, but with division protection. The variance value logVAR2 is used to calculate the VAREQ for the subsequent CSI and branch metrics after equalizer gain adjustment. Note that the last line of the above algorithm calculates vareq (col) in a manner that meets certain conditions. This is not just the inverse log calculation that converts logVAR (logarithm of variance estimate) to VAR. This accounts for the fact that the variance is calculated for the values that have not yet been equalized, and thus, adjustments are made to be compatible with the output equalized symbol values. In addition, the adjustment is also used to avoid variance estimation errors that occur in the presence of strong interference. These adjustments are included in the factor max [ Eqmagsq (col, max (Eqmagsq))/2 ] after the inverse logarithm.

"Algorithm for computing EQ and VAREQ from MED and logVAR, filter delay 16 symbols"

"IIR Filtering the MED and logVAR for each column to obtain MED1 and logVAR1, q 1/8IIR coef"

MED1＝(1-q)·MED1+q·MED；

logVAR1＝(1-q)·logVAR1+q·logVAR；

"smoothing MED1 and logVAR1 over columns using quadratic fit interpolation"

"MED 2 and logVAR2 are frequency smoothed mean and variance estimates"

MED2 ═ QF (MED 1); "calculating quadratic fit using QF Algorithm"

logVAR2 ═ QF (logVAR 1); "calculating quadratic fit using QF Algorithm"

"calculating equalizer coefficients EQ from MED 2"

medsq(col)＝|MED2(col)|²(ii) a 24 "save square amplitude"

24 "equalizer coefficients, T ═ training symbols"

"computing the inverse logarithmic sum of logVAR2 and equalizing to generate VAREQ"

EQmagsq(col)＝|EQ(col)|²；col＝0...24

VAREQ(col)＝2^logVAR2(col)·max[EQmagsq(col)，max(EQmagsq)/2]；col＝0...24

The EQ (col) values are then applied to the corresponding data-bearing symbols to generate OFDM EQ (col) values for each column of OFDM symbols (delayed by 22OFDM symbols to account for EQ processing delays). Vareq (col) is used for subsequent CSI processing.

Ofdm (col) · eq (col); 24 "equalizing delayed OFDM symbols"

The above algorithm uses a function called QF, which is a quadratic fit of the MED1 or logVAR1 matrices, for smoothing the values on the columns (subcarriers) of these matrices. Smoothing these values reduces estimation and correction errors due to noise, since it is assumed that the variations to be equalized will be smoothed. Variations in these values across the columns can be the result of several factors. One factor is the residual symbol tracking timing error due to the linear phase shift on the subcarriers. Since the filtering is done in the I and Q complex domain, and not phase and amplitude, the I and Q components resulting from this linear phase shift cannot be accurately corrected by linear fitting, while providing sufficient accuracy for quadratic fitting of the I and Q complex components. Another variation may be due to phase and amplitude perturbations due to frequency selective fading on the subcarriers, which may also be corrected by quadratic fitting. For phase and amplitude ripple from analog filtering prior to OFDM demodulation, if the ripple is small, it can be corrected. The interference also tends to have a logVAR shape that can be adapted to a quadratic fit.

If the analog filter is strongly pulsed and deviates from quadratic shape, a different QF function is required. Thus, two algorithm options are given: the first QF function is best suited to correct for variations due to residual symbol timing error, selective channel fading, and moderate filter ripple; the second algorithm is used to correct for all these variations and the more severe filter ripple.

On the subcarriers to which the quadratic shape is fitted to perform the smoothing correction, the first QF function estimates three points. At the midpoint and at the two extreme ends over the subcarrier range, FIR filters are used to estimate these points. The middle point is correctly estimated using a symmetric FIR filter on the middle subcarrier. The FIR filter at the endpoint has a centroid several bins from the endpoint. Although quadratic fitting is typically used to use the appropriate centroid close to the end points and extrapolate (extrapolation) the remaining subcarriers at the extreme end points, performance tends to be better if the centroid is assumed to be at the extreme subcarrier locations. The reason is that extrapolation tends to emphasize the curvature of the quadratic fit in the presence of noise. However, the algorithm may be modified to place the centroid at the location that yields the best overall performance.

Step 4 a: the first quadratic fit function smoothes the estimates in a partition shape (assumed to be quadratic) that provides near-optimal smoothing given possible channel conditions such as time offset and selective fading characteristics. This is achieved by using the following algorithm.

QF (x), quadratic fitting function, input vector x, and output vector y. (25 element vector) "

11 for k ═ 0.. 11; storage coefficient of filter point

"output value at col 0"

"output value at col 24"

"median value at col 12"

"second order coefficient a"

"second order coefficient b"

y(col)＝a·col²+ b. col + ylow; 24, col ═ 0.. 24; "output vector y"

An optional quadratic fit function QF is provided to accommodate IF filters with excessive ripple and group delay or gain variation. This function is different from the first function because different quadratic curves are used at each subcarrier location to form the FIR filter coefficients. These quadratic curves are pre-calculated and stored in a 25 x 25 matrix W to act as multipliers for the 25-value rows from the subcarriers to be filtered. Thus, rather than computing a quadratic fit over 25 subcarriers for each new OFDM symbol as in the first algorithm, the second algorithm simply multiplies a vector of 25 subcarrier values by the matrix W for each OFDM symbol time.

This optional QF function is applied in the same way as the method known under the name Savitsky-golay (sg) program; however, the alternative QF function generates the coefficients in a different way, which results in an improved anti-noise filtering gain while addressing the endpoint problem. The SG program calculates a least squares fit to each endpoint as the center to smooth the data for that point. The result is a set of FIR filter coefficients for each sub-carrier position to be smoothed. Two factors contribute to the use of least squares smoothing. One factor is the variability of the values over the subcarriers and another factor is the endpoint problem, i.e. the subcarriers close to the endpoint cannot be fitted by a symmetric set of FIR filter coefficients since no subcarriers are available for filtering outside the endpoint. The SG program generates FIR coefficients using a property that manipulates the Vandermonde matrix, thereby generating a unique set of FIR filter coefficients for each subcarrier location to be smoothed. Although the SG program generates FIR filter coefficients that result in an unbiased estimation for each smoothed subcarrier value, the actual set of FIR coefficients does not have the best noise reduction filtering characteristics due to the excessive use of negative coefficient values. However, the alternative QF function uses the best possible quadratic fit FIR coefficients for noise reduction filtering or smoothing while maintaining the zero-bias characteristic of the SG program. Furthermore, the alternative QF function has more flexibility in establishing the FIR filter smoothing range around the subcarrier locations.

One example of an alternative QF function is described below. The range of non-zero FIR filter coefficients for each subcarrier location is set to 15 non-zero coefficients to fit the expected variation of values across the 25 subcarriers in the partition, although it may also be adjusted. A unique FIR filter coefficient is calculated for each subcarrier location m 0.. 24. The shape of the FIR coefficients is a quadratic function with four additional constraints defined as:

constraint 1: the number of non-zero FIR filter coefficients is 15 and the remaining are 10 zero coefficients. The central non-zero coefficient is typically on the sub-carrier to be smoothed, which results in a symmetric FIR filter characteristic, except that for the 7 sub-carriers on either end, it is constrained by using the 15 sub-carrier positions on that end for the non-zero coefficients. The first non-zero coefficient position p for estimating (filtering) subcarrier m may then be identified by:

p ═ max (0, min (17, m-7)); "p is the first non-zero coefficient position".

Constraint 2: each of the 25 sets of 25 FIR coefficients (with 15 non-zero coefficients, 10 zero coefficients) must be added to unity value so that each FIR filter has a dc gain of 1 for each subcarrier location.

For the k coefficient estimated for the m subcarrier.

Constraint 3: the center of mass of the FIR filter for subcarrier m must also be m to ensure that an unbiased estimate is produced when the slope of the subcarrier data is assumed to be piecewise linear.

Constraint 4: although optimal noise reduction can be achieved by minimizing the sum of the squares of the coefficients, this does not provide the best local estimate for each subcarrier location and will result in 15 linear coefficients being generated for each FIR filter. A better constraint is to ensure that the quadratic function crosses zero at an unused coefficient position just outside the 15 non-zero coefficients. This is possible for 11 carrier positions 7 to 17, but the constraint is not met for other sub-carrier positions affected by the end point problem. Then the outer subcarrier locations have a zero crossing constraint only towards points that are outside the FIR coefficient range.

y_m(k)＝a_m·k²+b_m·k+c_m(ii) a "second order of k coefficient for m subcarrier"

y_m(p-1) ═ 0; "constraint for m 7.. 24"

y_m(p +15) ═ 0; "constraint for m 0.. 17"

Constraint 1 establishes only a range of 15 non-zero coefficients over which each of the 25 FIR filters has a quadratic characteristic. Constraint 2, 3 and 4 construction the quadratic coefficient a is determined for each FIR filter_m，b_mAnd c_mThree equations are necessary. While constraint 4 may appear to be overdetermined (over-determined) for the intermediate set of filter coefficients of m 7.. 17, with zero endpoints at both ends, the dual constraint for these subcarriers is redundant and all sets of coefficients are correctly determined. The algorithm for generating the optional QF1(x) FIR filter coefficient matrix W is next defined, with the resulting coefficient values for W as shown below.

And 4 b: an optional quadratic fit function QF is provided to fit IF filters with excessive ripple and group delay or gain variation. This function is different from the first function because the FIR filter coefficients are formed using different quadratic curves at each subcarrier location. These quadratic curves are pre-calculated and stored in a 25 x 25 matrix W to act as multipliers for the 25 value rows from the subcarriers to be filtered. Thus, rather than computing a quadratic fit over 25 subcarriers for each new OFDM symbol as in the first algorithm, the second algorithm simply multiplies a vector of 25 subcarrier values by the matrix W for each OFDM symbol time. These are subject to constraints 1-4, which results in the following algorithm.

"QF 1(x), optional quadratic fit matrix function, input row vector x, output vector y. "

"first calculation pre-stored coefficient matrix W (25 × 25)"

FORm＝0to7

c(m)＝-225·a(m)-15·b(m)

FORk＝0to14

W(k，m)＝a(m)·k²+b(m)·k+c(m)

W(24-k，24-m)＝W(k，m)

FORm＝8to16

FORk＝0to14

W(k+m-7，m)＝W(k，7)

"Up to this point, the pre-storage calculation for the filter matrix W is finished"

"calculating the filtered output vector y for each new OFDM symbol"

y is x.W; matrix multiplication to generate output vector y "

And 4 c: a third alternative quadratic fit is described below.

Another optional filter QF2(x) may be designed using all 25 possible non-zero coefficients for each FIR filter. This filter has characteristics more similar to the first qf (x) filter, but is constructed using a matrix form W of selectable filters.

"QF 2(x), optional quadratic fit matrix function, input row vector x, output vector y. "

"first calculation pre-stored coefficient matrix W (25 × 25)"

FORm＝0to13

FORk＝0to24

W(k，m)＝a(m)·k²+b(m)·k+c(m)

W(24-k，24-m)＝W(k，m)

"Up to this point, the pre-storage calculation for the filter matrix W is finished"

"calculating the filtered output vector y for each new OFDM symbol"

y is x.W; matrix multiplication to generate output row vector y "

In another aspect, the invention includes adaptive complementary combining of second level partitions prior to equalization. The two independent second level partitions are independently equalized and correlated VAREQ estimation is performed. The computation of the branch metrics is independent and redundant for all second level soft code bits in the partition. The respective branch metrics are then added to generate a set of branch metrics. In addition, equalization is performed on the complementary combined second level partitions to generate another set of branch metrics for the same set of second level soft code bits. Then, for each second level soft code bit, a higher branch metric is selected as an output for the corresponding second level soft code bit.

The use of complementary subcarriers for the hybrid second and third level partitions creates an orthogonal relationship with its analog host. Existing implementations of second-level equalization require knowledge of whether to limit the analog host bandwidth to ± 5 kHz. If the analog bandwidth is limited to 5kHz, the second stage partitions are equalized independently to better account for adjacent channel interference. Otherwise, the second level partitions are first complementarily combined to eliminate the analog signal in that region.

The input symbols to be equalized are delayed to match the delays in the equalizer parameter estimates and to provide timely application of the equalizer information. The EQ (col) values are then applied to the respective data-bearing symbols to generate an OFDM EQ (col) value for each column of OFDM symbols (delayed by 22OFDM symbols to account for EQ processing delays). The vareq (col) value is used for the subsequent CSI processing.

The method of the invention does not use analog bandwidth information; but rather performs independent and combined equalization, with the largest branch metric being selected later. This makes the performance more robust, especially when the analog bandwidth is somewhat beyond 5 kHz.

The third stage subcarriers are always combined complementarily prior to equalization. A third level of equalization is then performed as described. The two second level partitions are processed independently and in complementary combination, generating three sets of equalized branch metrics for a single set of second level soft code bits. The method of combining these three sets of branch metrics is described below.

The two independent second level partitions and the associated VAREQ estimates are independently equalized. Branch metrics are independently and redundantly computed for all second level soft code bits in the partition. The respective branch metrics are then added to generate a set of branch metrics. In addition, equalization is performed on the complementary combined second level partitions to generate another set of branch metrics for the same set of second level soft code bits. Then, for each second level soft code bit, a higher branch metric is selected as the output of the corresponding second level soft code bit.

As described above, the equalizer includes two parts: flat fading compensation (equalizer) followed by a partitioned equalizer. In fast fading scenarios, a flat fading equalizer is advantageous, which uses the main carrier (FFT bin 0) and BPSK subcarriers (bins ± 1). The partition equalizer is slower and operates on training symbols that are sparser in the partition, but more accurate in the partition. The partitioned equalizer benefits from a flat fading equalizer to keep the training values within a relatively smaller range.

The functions illustrated in the figures may be implemented using known circuit elements, including but not limited to one or more processors or application specific integrated circuits.

Although the invention has been described in terms of several examples, those skilled in the art will recognize that many changes may be made to the examples without departing from the scope of the invention, which is described in the claims that follow.

Claims (3)

1. A method of equalizing an OFDM symbol vector received on an AM in-band on-channel radio signal including a main carrier and first and second BPSK modulated subcarriers adjacent to the main carrier, the method comprising the steps of:

calculating a BPSK magnitude signal;

filtering the BPSK magnitude signal;

filtering the complex samples received on the main carrier;

calculating a plurality of flat fading equalization coefficients using the filtered amplitude of the BPSK signal on the first and second BPSK modulated subcarriers and the filtered complex samples received on the main carrier; and

multiplying the OFDM symbol vector by the plurality of flat fading equalization coefficients.

2. The method of claim 1, wherein the step of filtering the BPSK magnitude signal comprises the steps of:

the BPSK magnitude signal is passed through a median filter and a finite impulse response filter.

3. A receiver for receiving an AM in-band on-channel radio signal including a main carrier and first and second BPSK modulated subcarriers adjacent to the main carrier, the receiver comprising:

an input for receiving an AM in-band on-channel radio signal;

an equalizer for calculating a BPSK magnitude signal, for filtering the BPSK magnitude signal, for filtering complex samples received on the main carrier, for calculating a plurality of flat-fading equalization coefficients using the filtered magnitudes of the BPSK signal on the first and second BPSK modulated subcarriers and the filtered complex samples received on the main carrier, and for multiplying an OFDM symbol vector by the plurality of flat-fading equalization coefficients; and

an output device for generating an output in response to the AM in-band on-channel radio signal.