JP2008153751A - Normalization method and receiver of soft decision signal - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a normalization method and a receiver of soft decision signals capable of achieving high reception performance with a small number of bits. <P>SOLUTION: The normalization method comprises: a demodulation means 11 for performing a soft judgement demodulation of reception signals; a mode value detection means 17 for detecting a mode value of signal amplitude based on frequency distribution of amplitude of a soft judgement signal sequence Lj between decoding unit sections of output of the demodulation means; a normalization means 13 for normalizing the amplitude of the soft judgement signal sequence Lj with the detected mode value as a reference; and a decoding means 14 for decoding reception data based on the soft judgement signal sequence after the normalization. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a soft decision signal normalization method and a receiving apparatus, and more particularly to a soft decision signal normalization method and a receiving apparatus for normalizing a soft decision signal output from a demodulator and applying it to a decoder. The present invention is suitable for application to a receiving apparatus that performs soft decision error correction decoding under a multi-level modulation scheme such as QAM in a mobile communication system.

  A receiving apparatus that performs error correction decoding based on a soft decision signal requires a wide dynamic range for signal processing in order to adopt multi-level QAM and adapt to a severe fading environment. The expansion of the dynamic range leads to the expansion of the bit accuracy, and the circuit scale increases.

  FIG. 9 is a diagram showing an example of the multi-level modulation method, and shows the concept of 16QAM demodulation. In 16QAM, 4-bit data “0000” to “1111” constituting a transmission symbol (code point) is encoded by being divided into two groups of bits b0 and b2 and bits b1 and b3, and transmitted and demodulated. Soft-decision demodulation at level 1 includes level 1 soft-decision representing the distance from the I and Q axes, and i. The determination is performed in two steps: a soft decision of level 2 representing a distance from the q axis (where the i.q axis is determined from the average amplitude value of the received I and Q signals). For example, when the soft decision demodulation of code point “0000” is viewed on the I axis, the soft decision output at level 1 is “3a”, the soft decision output at level 2 is “a”, and the soft decision output at code point “0001” is When the soft decision demodulation is viewed on the I axis, the soft decision output at level 1 is [a] and the soft decision output at level 2 is also [a].

  FIG. 10 shows likelihood histograms in various multilevel modulation schemes. The horizontal axis is a log-likelihood ratio (LRR) with a base of 2, and the vertical axis is a relative frequency. Each graph has a typical shape corresponding to various modulation schemes under high SNR conditions. In the case of QPSK in FIG. 10 (A), only the determination of level 1 is performed. Therefore, when the absolute value of likelihood is viewed on the I (or Q) axis, the distribution is due to noise or the like with a single peak as the center. It becomes something with a spread. In this case, the average likelihood Av is located at the approximate center of the distribution. In the case of 16QAM in FIG. 10B, two determinations of level 1 and level 2 are performed, so that two peaks appear in the likelihood distribution. In the determination of level 1, the determination of amplitudes a and 3a occurs at a ratio of 1/2, and in the determination of level 2, all are determinations of amplitude a. Therefore, the generation frequency of amplitude a is 75% and the generation frequency of amplitude 3a. Is a ratio of 25%. In this case, the average value Av is located approximately in the middle of the two peaks. Similarly, in the case of 64QAM, the distribution is as shown in FIG. 9C, and the average value Av is located approximately in the middle of these peaks.

  Therefore, the dynamic range necessary to cover these likelihood (amplitude) distributions has not only the spread due to fading and noise, but also the spread inherent to various modulation schemes, and the wider the number of multi-values, the wider A dynamic range is required. Also, the positional relationship between the likelihood distribution and the average value varies depending on the number of multivalues. Therefore, if such a soft decision demodulated output is input to the decoder as it is, the scale of the arithmetic circuit and the memory becomes large, so that the dynamic range is narrowed down by normalizing the soft decision signal and reducing the number of bits. .

  Conventionally, as shown in Patent Documents 1 and 2, a reference value having a fixed relationship with the average value is created based on an average value of an input soft decision signal sequence, and the soft decision signal sequence is generated using the reference value. It was common to divide (normalize). This will be specifically described below.

FIG. 11 is a diagram illustrating a main configuration of a conventional receiving apparatus. In the figure, an input A / D converted received signal is soft-decision demodulated by a demodulation circuit 51, and an output soft-decision signal sequence L _j (j = 1, 2).
,... J) are temporarily stored in the memory 52. On the other hand, these soft decision signals Lj are converted into absolute values by the absolute value circuit 55, and the average value X of the amplitude of the soft decision signal sequence in the decoding unit section (code block unit or frame unit) J is further obtained by the average value detection circuit 56. Desired. The reference value creating circuit 57 divides this average value X by a predetermined set value A to create a reference value B (= X / A). Further, the normalizing circuit 53 reads each soft decision signal L read from the memory 52. Normalize by dividing _j by the reference value B. An FEC (Forward Error Correction) decoding circuit 54 performs error correction decoding on the received data based on the normalized soft decision signal (= L _j / B).
JP 2004-260713 A Japanese Patent Laid-Open No. 2002-232302

  However, even if the modulation method is fixed, the average value fluctuates not a little due to the influence of amplitude fluctuation due to severe multipath fading. The influence due to the amplitude fluctuation becomes significant when the moving body moves at a high speed and the fluctuation cycle becomes relatively short with respect to the decoding unit section. In recent years, wireless signals have become wider in bandwidth, and in this case, severe amplitude fluctuations also occur due to frequency selective fading due to multipath.

  Under such circumstances, we want to put as many demodulated signals as possible into the decoder (dynamic range) based on the average value. However, since the average value fluctuates greatly due to the influence of fading, it falls within the dynamic range. A lot of no signal comes out. Although there is a method of sliding a certain dynamic range up / down, it is difficult to cover the lower limit side of the input signal in particular because the average value cannot capture the variation pattern of the likelihood distribution.

  In this way, since the average value may not capture the characteristics of the likelihood distribution, if the frequency distribution of the signal widens due to the modulation method and fading, the average value becomes the peak of the frequency distribution. Therefore, the signal amplitude of the normalized result cut out within the limited number of bits becomes “0”, or the likelihood at the upper limit of the amplitude is clipped, and the determination value is stuck.

  When the frequency at which the amplitude becomes “0” is large, it is an increase in the indeterminable signal, which means that the signal is not transmitted. Accordingly, the decoding performance is greatly deteriorated. On the other hand, when the frequency of clipping is high, the state is close to a hard decision, but for the decoding circuit, only the SN ratio is degraded and the influence is small. On the other hand, the generation of a signal with an amplitude of “0” not only deteriorates the SN ratio, but also increases the coding rate, and in the case of systematic codes, causes systematic bits to be lost, and has a large effect on performance degradation. As described above, in normalization based on the average value, normalization outside the important signal range may be performed, and reception performance may be significantly deteriorated due to this loss.

  The present invention has been made in view of the above-described problems of the prior art, and an object of the present invention is to provide a soft decision signal normalization method and a receiving apparatus that can obtain high reception performance with a small number of bits. is there.

  A soft decision signal normalization method according to a first aspect of the present invention is a soft decision signal normalization method for normalizing a soft decision signal output from a demodulator, and includes a frequency distribution of soft decision signal amplitudes in a decoding unit interval. The mode value of the signal amplitude is detected based on the above, and the soft decision signal sequence is normalized based on the detected mode value.

  FIG. 10 shows a histogram of likelihood distribution in the multi-level modulation method. As described above, the average value Av is a relatively good index for evaluating the amplitude fluctuation (dynamic range) of the received signal. However, the average value Av may vary between each amplitude distribution pattern by the modulation method (multi-level number) or multipath fading. Since the positional relationship of these changes considerably, it is not necessarily an index that well reflects the reception environment.

  In this respect, as shown in the figure, the mode Mo is always settled at the highest amplitude frequency regardless of the modulation method, so that it is not affected by the difference in the modulation method (multi-value number), and multipath fading. Even if the reception amplitude largely fluctuates due to the above, etc., the mode value (large number) of the reception level is always well reflected.

  In general, the decoding gain due to the soft decision signal greatly depends on the level resolution near the origin of the soft decision signal. For example, the mode value Mo of the QAM method always appears at a low amplitude, and based on this, the soft input signal is softened. Even if the determination signal is normalized, a relatively high level resolution can be maintained near the origin. On the other hand, as a result of giving priority to the lower level resolution, even if the higher signal level is clipped, the influence of the normalized soft decision signal on the decoding gain is small. Therefore, a method of suppressing the generation of 0 signal while allowing a certain degree of degradation due to clipping is effective. By using the mode value for level adjustment of the dynamic range according to the present invention, reception quality can be greatly improved, The number of bits of the soft decision signal after normalization can be reduced.

  In the second aspect of the present invention, the soft decision signal amplitudes composed of binary numbers are classified in units of powers of 2, and the maximum distribution is included based on the frequency distribution of the soft decision signal amplitudes included in each class. The power value is the mode value. Therefore, the mode value can be detected efficiently with a simple configuration and processing.

A receiving apparatus according to a third aspect of the present invention includes a demodulating means for soft-decision demodulating a received signal;
A mode value detecting means for detecting a mode value of the signal amplitude based on a frequency distribution of the soft decision signal amplitude in the decoding unit section of the output of the demodulating means; and a mode value of the soft decision signal sequence based on the detected mode value The image processing apparatus includes a normalization unit that performs normalization and a decoding unit that decodes received data based on the normalized soft decision signal sequence.

  In the fourth aspect of the present invention, the mode detection means classifies each soft decision signal amplitude consisting of binary numbers in units of powers of 2, and based on the frequency distribution of the soft decision signal amplitude included in each class. The power of 2 of the class including the maximum distribution is defined as the mode value.

In the fifth aspect of the present invention, the mode detection means includes a plurality of counters provided for each class in units of powers of 2 and a control means for performing count control of each counter.
The control means sets each counter to 0 or the maximum value at the initial stage of the decoding unit interval and monitors the subsequent count output, and the soft decision signal amplitude is included with the input of each soft decision signal. If only the counter corresponding to each power of 2 of the class that did not exist is counted up or down, and the count output of all counters is not 0 or the maximum value, the count output of any counter becomes 0 or the maximum value Until all the counters are counted, the control for counting down or counting up is repeated, and the power of 2 corresponding to the counter whose count output is 0 or the maximum value at the end of the decoding unit interval is set as the mode value. Therefore, the mode value can be efficiently detected with a simple hardware configuration and control.

  In the sixth aspect of the present invention, the normalizing means divides the soft decision signal amplitude consisting of the binary number of the demodulator output by a reference value consisting of a power of 2 smaller than the mode value. The division of the binary soft decision signal by a power of 2 can be easily performed by bit-shifting the soft decision signal to the lower side.

  In the seventh aspect of the present invention, when the number of bits of the soft decision signal output from the demodulator is N and the number of bits of the soft decision signal after normalization is M, the reference value is 0 to (N−M). Restricting means for limiting within the range is provided.

  In the present invention, even when the mode value is small, the reference value is limited to 0 (no bit shift) or more, so that the lower bits of the input soft decision signal can be effectively used as they are for soft decision decoding. On the other hand, even when the mode value is large, the reference value is limited to (N−M) or less, so that it is effective for soft decision decoding without losing the lower bits of the input soft decision signal more than necessary (without shifting out). Available to: Also, M bits can always be secured in the normalized soft decision signal.

  According to an eighth aspect of the present invention, there is provided a receiving device for detecting a mode value based on a frequency distribution of a soft decision signal amplitude in a decoding unit section of an output of the demodulation unit and a demodulation unit for soft decision demodulation of a received signal. A value detecting means, a normalizing means for normalizing the soft decision signal amplitude based on a reference value based on the detected mode value, and a soft decision after normalization relating to at least an initial transmission packet under packet retransmission control A memory for storing a signal sequence and a reference value, a combining means for combining a maximum ratio of each normalized soft decision signal amplitude relating to an initial transmission packet and a retransmission packet based on a ratio of the respective reference values; Decoding means for decoding received data based on the soft decision signal sequence.

  In the present invention, under the packet retransmission combining method (HARQ), the normalized soft decision signal amplitudes of the initial transmission packet and the retransmission packet are maximized based on the ratio of the reference values according to the respective mode values. Since the ratio combining is performed, the optimum maximum ratio combining can always be performed according to the reception environment such as multipath fading regardless of the multi-level modulation method.

  According to a ninth aspect of the present invention, there is provided a receiving apparatus for demodulating a received signal soft decision demodulating and parallel converting a serial soft decision signal sequence in a decoding unit section of the output of the demodulating unit in units of a predetermined number of symbols. Serial-parallel conversion means, and each of the parallel-converted soft decision signal sequences provided in parallel to classify the amplitude of each soft decision signal sequence consisting of binary numbers in units of powers of 2, A plurality of distribution counting means for counting the frequency distribution of the soft decision signal amplitude included in the class, an adding means for adding the count output for each class by each distribution counting means by class, and a maximum based on the frequency distribution of the added output A mode value detecting unit that uses a power of 2 of a class including the distribution of the mode as a mode value, a normalizing unit that normalizes the soft decision signal amplitude based on the mode value, and the normal In which and a decoding means for decoding the received data based on the soft decision signal sequence after. Therefore, the mode value can be detected at a high speed, and it is possible to cope with broadband communication.

  As described above, according to the present invention, a high reception performance can be obtained with a smaller number of bits, so that it greatly contributes particularly to downsizing and performance improvement of mobile devices.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of the invention will be described in detail with reference to the accompanying drawings. Note that the same reference numerals denote the same or corresponding parts throughout the drawings.

  FIG. 1 is a configuration diagram of a main part of a receiving apparatus according to the first embodiment. A reference value is created from a mode value of an amplitude (likelihood) distribution obtained for a soft decision signal sequence in a decoding unit section. Shows the case where each soft decision signal is normalized.

An outline of the operation will be described. In the figure, the demodulation circuit 11 soft-demodulates the input A / D converted received signal by 16QAM, 64QAM or the like, and the memory 12 in the decoding unit section (code block unit or frame unit). Each soft decision signal L _j (j = 1, 2,... J) output from the demodulation circuit is temporarily stored for time adjustment. On the other hand, the absolute value circuit 15 converts the amplitude (sign) of each soft decision signal L _j into an absolute value, and the indexing circuit 16 exponentiates the amplitude (likelihood) of each soft decision signal Lj in units of powers of 2. (Classification). Further, the mode value detection circuit 17 counts the frequency of the amplitude included in each class with respect to the amplitude of each soft decision signal Lj after indexing, and sets the power value P of 2 representing the class with the highest frequency as the mode value. Output. Further, the reference value creating circuit 18 divides (subtracts) the mode value P by a predetermined set value A to create a reference value B (= P−A), and the limiting circuit 19 sets the reference value B to a predetermined value. Limit so as not to depart from the scope. On the other hand, the normalization circuit 13 divides each soft decision signal L _j read from the memory 12 by the reference value B to normalize. An FEC (Forward Error Control) decoding circuit 14 performs error correction decoding of the received data based on each soft decision signal (= L _j / B) after normalization.

  FIG. 2 is a diagram for explaining the operation of the absolute value and indexing circuit according to the embodiment. The operation when these functional blocks are configured by hardware or a DSP will be specifically described. In the absolute value circuit 15, for example, when a positive soft decision signal “39” is input, the data portion is output as it is, and when a negative soft decision signal “−39” is input, the two's complement of the data portion is output. The absolute value is obtained and the obtained soft decision signal “39” is output. In the case of hardware, a 1's complement (bit inversion signal) “38” of the data part is generated, and “1” is added to this to generate a 2's complement “39”.

The indexing circuit 16 obtains a logarithm log ₂ “39” = 5.29 with 2 as the base for the absolute value “39” of the soft decision signal Lj, and extracts only the integer part “5” to obtain the amplitude. Index (likelihood). In the case of hardware, the bit “1” is searched in order from the most significant bit b7 of the register holding the absolute value “39” binarized, and the bit b5 = “1” found first is left, The lower bits are set to “0”. In this case, 2 ⁵ (= 32) is the most significant bit. At this time, since a fractional value less than ²⁵ is rounded down, bit b5 = "1" is shifted to the upper side by 1 bit, and bit b6 = "1" is set as a class (exponent) representing the absolute value "39". . In this example, the absolute value “39” is classified into the range of 2 ⁵ ≦ “39” <2 ⁶ to ²⁶ .

Generalizing this, the absolute value Lj included in the scope of 2 ^n-1 ≦ absolute value Lj <2 ⁿ to class divided into 2 ^n, thus for example, all of the input signal composed of 8 bits, 2 ¹ , 2 ² , 2 ³ ,..., 2 ⁸ are indexed (classified). In this embodiment, classification is performed by binary digits (that is, exponents), so that a wide range can be expressed by a small number of classes, thereby reducing the circuit scale. The absolute value Lj included in the range of 2 ⁿ⁻¹ ≦ absolute value Lj <2 ⁿ may be classified into 2 ⁿ⁻¹ .

  FIG. 3 is a block diagram of the mode detection circuit according to the embodiment. In this embodiment, the mode value is detected with a small amount of hardware in association with the indexing of the input soft decision signal amplitude. We propose a case where it can be done efficiently.

In the figure, K-bit up / down counters CTR0 to CTR7 are connected to the bits b0 to b7 (when N = 8) of the output of the indexing circuit 16, respectively. If the number of data in the normalization interval is J, the number of bits k of each counter is
K> J / 2 where K = 2 ^k
It is good to satisfy. The reason will be described later. Each counter CTR outputs a logic 1 level (eg, high level) of the carry-out signal Co when the output count value Q = (K−1), and outputs a logic 1 level of the borrow signal Bo when the count value Q = 0. Output.

The carry-out signals C0 to C7 of the counters CTR0 to CTR7 are input to the NOR gate circuit NO1, and when all the carryout signals C0 to C7 are at logic 0 level (low level), the output of NOR1 is energized. All counters CTR to CTR7 can be counted up. In addition, the inverted signals of the borrow signals Bo of the counters CTR0 to CTR7 are respectively input to the AND gate circuits A0 to A7, and the counter whose count output Q is 0 is the input of the corresponding AND gate circuit. Will no longer count down.

  From the viewpoint of hardware reduction, it is desirable to reduce the number of bits of the counters CTR0 to CTR7 as much as possible. If a certain counter is continuously counted down by J soft decision signals, the counter needs to be able to count down to a maximum of J. However, in this embodiment for the purpose of detecting the mode value, even if the counter counts down to at least J / 2 even in such a situation, this counter can be estimated as one mode value candidate at this point.

  Because, depending on the signal amplitude distribution, another counter may count down to the minimum J / 2 at the same time. In this case, the other counter can be estimated as another candidate of the mode value, In this case, the power value of 2 of the counter corresponding to the small amplitude value can be determined as the mode value P. This is the worst case, and it is not possible for three or more counters to count down to J / 2 at the same time, so each counter may be capable of counting up to J / 2.

  With such a configuration, the maximum value I (= K−1) is set as an initial value in CTR0 to CTR7 by the pulse FP generated at the beginning of the decoding period (normalization period), and thereby the carry signals C0 to CTR0 of the counters CTR0 to CTR7 are set. C7 is all at a logic 1 level. Next, when the first exponential likelihoods b0 to b7 are input, only the counter corresponding to the input of the bit “0” is counted down in synchronization with the timing pulse TP generated at that time. For example, when only the bit b6 of the exponential likelihood is a logic 1 level, only the other CTR0 to CTR5 and CTR7 are counted down by the clock signal CK. Next, when only the bit b5 of the exponential likelihood is at the logic 1 level, only the other CTR0 to CTR4 and CTR6 and 7 are counted down.

  At this time, the carry signals C0 to C7 of all the counters CTR0 to CTR7 are all at the logic 0 level, and as a result, the output of NOR1 is at the logic 1 level, so that all the counters CTR0 to CTR7 are simultaneously counted up by the clock signal CK. To do. As a result, when the carry-out signal Co of any counter becomes the logic 1 level, the output of the NOR gate circuit NO1 is deactivated and the count-up is stopped.

  Thus, when J soft decision signals in the normalization interval are input, only the carry-out signal Co of the counter connected to the indexing bit having the largest logical 1 level remains at the logical 1 level. In this way, the mode value P (corresponding to 2 to the power of P) of the exponential amplitude can be efficiently detected with a small amount of hardware. Actually, there may be cases where the carry-out signals Co of a plurality of counters are both at a logic 1 level. In this case, from the viewpoint of respecting the normalization accuracy of the relatively low level soft decision signal, the smaller exponent bit is set as the mode value.

In the above embodiment, the mode detection means includes a plurality of counters provided for each class whose unit is a power of 2, and a control means for performing count control of each counter, wherein the control means includes a decoding unit. At the initial stage of the section, each counter is set to the maximum value and the subsequent count output is monitored, and with the input of each soft decision signal, the soft decision signal amplitude is not included in each power of 2 If only the corresponding counter is counted down and the count output of all counters is not at the maximum value, repeat the control to count up all the counters simultaneously until the count output of one of the counters reaches the maximum value. Although the case where the power value of 2 corresponding to the counter having the maximum count output at the end of the interval is set as the mode value has been described, the present invention is not limited to this.

  In addition, the control means sets each counter to 0 at the initial stage of the decoding unit interval and monitors the subsequent count output, and the soft decision signal amplitude is included as each soft decision signal is input. If only the counter corresponding to each power of 2 of the class that has not been counted is counted up and the count output of all counters is not 0, all the counters are all at once until the count output of any counter becomes 0 The countdown control may be repeated so that the power value of 2 corresponding to the counter whose count output is 0 at the end of the decoding unit interval is set as the mode value.

  FIG. 4 is a diagram for explaining the operation of the reference value creation and normalization circuit according to the embodiment. By the way, if the input soft decision signal Lj is normalized (divided) by the mode value P as it is, the amplitude of the soft decision signal distributed below the mode value becomes “0”. It is necessary to set a smaller value as the reference value B. Therefore, the reference value creation circuit 18 creates a reference value B (= P−A) by subtracting a predetermined set value A from the mode value P. The set value A is a constant (a logarithm with 2 as the base) for covering the lower distribution of the mode P, and usually a soft decision gain can be secured by setting it to about “4” or “5”.

  In FIG. 4A, when the reference value B is simply obtained from the mode value P by the above method, when the mode value P (= 8), the set value A (= 5) bits below it. If a soft decision gain is to be secured, the reference value B (= 8−5 = 3) is obtained. Further, when the mode value P (= 7), the reference value B (= 2) is obtained. Similarly, when the mode value P (= 4), the reference value B (= −1) is obtained.

By the way, when the input 8-bit soft decision signal is shifted by 3 bits to the lower side, the number of significant digits is “5”, so that M (= 6) bits cannot be provided to the decoding circuit 14. On the other hand, when the 5-bit soft decision signal is shifted to the lower side, the lower significant bits are discarded, so it is better not to shift the bit. For this reason, the limiting circuit 19 sets the reference value B within the following range:
0 ≦ B ≦ (N−M)
It is restricted to the ones. In the example of the figure, the range is limited to 0 ≦ B ≦ 2.

  In FIG. 4B, the normalization circuit 13 normalizes the input soft decision signal Lj by dividing it by the reference value B. Normalization by the reference value B (= 2) can be easily performed by shifting the input soft decision signal Lj by 2 bits to the lower side in hardware. Normalization by the reference value B (= 0) extracts the lower 6 bits as they are without shifting the input soft decision signal Lj to the lower side.

  FIG. 5 is a diagram for explaining the effect of the normalization method according to the embodiment. A simulation result when a signal subjected to multipath fading is received over a plurality of (for example, 1000) frames and soft decision demodulation is performed with 64QAM. Is shown. FIG. 5A shows a signal distribution before normalization, a characteristic Lj is a soft decision signal of all inputs, a characteristic Av is an average value in each decoding unit section, and a characteristic Mo is a relative frequency of a mode value in the same section. Respectively. Each distribution has a spread due to the influence of multipath fading, and the average value distribution Av is shifted to a larger amplitude than the center of the input signal distribution Lj by adopting 64QAM. On the other hand, the mode distribution Mo overlaps with the input signal distribution Lj and is distributed at a value lower than the average value distribution Av.

FIG. 5B shows an amplitude (likelihood) distribution after normalizing the input soft decision signal sequence with reference to the average value and the mode value, both securing several bits on the lower side. . In the figure, characteristic Lj
Indicates the signal distribution when the characteristic Lj / Av is normalized based on the average value, and the characteristic Lj / Mo indicates the signal distribution when normalized based on the mode value.

  In practice, there was no input signal having an amplitude of “0”, but there are not a few signals that have been included in the exponential likelihood “0” by normalization, and the frequency is an average value. The normalization with the mode value Mo is clearly less than the normalization with Av. Therefore, the normalization with the mode Mo is better than the normalization with the average value Av. On the other hand, in the higher level, normalization by the mode value Mo is distributed in a narrower range, but this sticking does not cause a large deterioration of soft decision decoding. Thus, in the conventional method of normalizing based on the average value, 8 to 10 bits are required for the soft decision signal after normalization, but in the present embodiment, this can be reduced to 6 bits.

  FIG. 6 is a graph of the block error rate according to the embodiment, and shows a case where the coding rate is 0.9 and the propagation condition is TU (Typical case for Urban area) 120 km / h by 64QAM demodulation. The horizontal axis represents the signal-to-noise ratio SNR, and the vertical axis represents the block error rate (BLER). A characteristic O represents the case of an ideal calculation obtained by decoding the input soft decision signal without normalization. The characteristic A1 is normalization based on the average value Av, and shows a case where the number of bits after normalization M = 7 bits and the set value A = 5. The characteristic A2 is also normalized based on the average value Av, and shows the case where the number of bits after normalization M = 8 bits and the set value A = 6. The characteristic M1 is normalized based on the mode value Mo, and shows the case where the number of bits after normalization M = 7 bits and the set value A = 5. The characteristic M2 is also the number of bits after normalization M = The case of 8 bits and setting value A = 6 is shown. The TU condition is described in the literature (3Gpp Technical Specification TS45.005 “Radio transmission and reception”).

  As can be seen from the graph, the BLER characteristic is better as the number of bits M (or set value A) after normalization is larger, and normalization based on the mode value Mo is performed rather than normalization based on the average value Av. BLER characteristics are better.

  FIG. 7 is a block diagram of the main part of the receiving apparatus according to the second embodiment, and shows an application example to the receiving apparatus using the packet retransmission combining method HARQ (Hybrid Automatic Repeat Request) of the present invention. The error correction technology is roughly divided into two types: a retransmission combining (ARQ: Automatic repeat request) method and a FEC (Forward error Control) method. There is also a HARQ (Hybrid ARQ) method that combines the advantages of both. Here, it is applied to a receiving apparatus that adopts a hybrid retransmission combining method (HARQ) that uses a maximum a posteriori probability (MAP) decoding method that is adopted in recent mobile radio to minimize a bit error rate after decoding. An example is shown.

  In the figure, the configurations of the demodulating circuit 11 to the normalizing circuit 13 and the FEC decoding circuit 14 may be the same as those described in FIG. The remaining part is related to HARQ control, and the operation of this part will be described below. In the first reception, a series of soft decision signals L1 normalized by the reference value B1 are input to the combined amplitude normalization circuit 21 together with the first reference value B1. At this time, since there is no past soft decision signal sequence to be combined, the series of soft decision signal sequences L1 is added to the FEC decoding circuit 14 as it is through the adder circuit 23 and stored in the memory 22 at the same time. The first reference value B1 is stored in the reference value memory 24. The FEC decoding circuit 14 performs error correction decoding based on the first soft decision signal Ll, and the decoded data of the output is CRC checked by a CRC check circuit (not shown). In the case of check OK, an ACK is returned to the transmitting side. To do. Then, the reception of this frame ends.

However, if the CRC check is NG, a NACK is returned to the transmitting side, and the transmitting side that receives this transmits a second frame for the same data. Further, on the receiving side receiving this, the second soft decision signal sequence L2 normalized by the reference value B2 is input to the combined amplitude normalization circuit 21 together with the reference value B2. At the same time, the first soft decision signal L 1 is read from the memory 22, and the first reference value B 1 is read from the reference value memory 24 and input to the combined amplitude normalization circuit 21.

  These soft decision signal sequences L1 and L2 are normalized (leveled) by the combined amplitude normalization circuit 21 to an amplitude corresponding to the ratio of the reference values B1 and B2, and then added by the adder 23 (maximum ratio combining). Are added to the FEC decoding circuit 14, and the soft decision signal sequence L2 and the reference value B2 are also stored in the memories 22 and 24, respectively. The FEC decoding circuit 14 performs error correction decoding based on the combined soft decision signal sequence output from the adder, and the output decoded data undergoes CRC check. In the case of check OK, an ACK is returned to the transmitting side. Then, the reception of this frame ends. In the case of inspection NG, NACK is returned to the transmission side, and the third soft decision signal L3 is synthesized.

  In the second embodiment, by using a reference value based on the mode value as a level matching reference, it is possible to properly normalize the amplitude of the soft decision signal under various reception environments. In the second embodiment, the mode value generated at the normalization stage can be effectively used for the maximum ratio combining process.

  FIG. 8 is a block diagram of the main part of the receiving apparatus according to the third embodiment, and shows a case where the mode value is detected at high speed. In the figure, the soft decision output of the demodulation circuit 12 is converted into a parallel signal for every N bits by a serial / parallel converter (S / P) 31, and one of them is stored in four memories 12a to 12d for time adjustment. The other is input to the four absolute value circuits 15a to 15d.

  In the present embodiment, the N (for example, 8) bit soft decision signal sequence Lj is subjected to S / P conversion for every 8 bits, so that the absolute value circuit 15a to the distribution counting circuit 32a in the first row are the first, fifth. , 9,..., The exponential likelihoods (amplitudes) b0 to b7 for the sub soft decision signal series are counted by class, and the absolute value circuit 15b to the distribution counting circuit 32b in the second row are the second, sixth, The indexing likelihoods b0 to b7 for the 10th,..., Sub-soft decision signal sequence are counted by class. The same applies to the following.

  Thus, when J (for example, about 5000 to 10000 symbols) soft decision signal sequences Lj are finally input, each of the distribution counting circuits 32a to 32d has J / 4 interleaved sub soft decision signal sequences. Indexing likelihoods b0 to b7 are counted individually. Therefore, the adding circuit 33 adds the count outputs B0 to B7 of the distribution counting circuits 32a to 32d for each class, and thus each of the exponential likelihoods b0 to b7 equivalent to that obtained for the J soft decision signal sequences. Output the number. The mode value detection circuit 34 detects an exponential likelihood (mode value) P having a large likelihood value based on the count output of the adder output. The reference value creating circuit 18 creates a reference value B based on the mode value P, and the limiting circuit 19 limits the reference value B to a value within a predetermined range.

  On the other hand, the normalization circuits 13a to 13d normalize the interleaved sub soft decision signal sequences with the reference value B to generate each M (for example, 6) bits normalized sub soft decision signal sequences. The parallel-serial converter (P / S) 35 deinterleaves each normalized sub soft decision signal sequence to convert it into a serial normalized soft decision signal sequence, and inputs it to the FEC decoding circuit 14.

  In the third embodiment, the input soft decision signal sequence Lj is decomposed into four sub soft decision signal sequences. However, the present invention is not limited to this. In addition, it can be decomposed into an arbitrary number of sub soft decision signal sequences.

  In each of the above embodiments, the mode value is detected by classifying the amplitude (likelihood) range of the soft decision signal sequence in units of powers of 2, but the present invention is not limited to this. The mode may be configured such that the mode value is detected by classifying the amplitude range into units of linear sections such as 10 or 100.

  Moreover, although the several embodiment suitable for the said invention was described, it cannot be overemphasized that the structure of each part, control, a process, and these combination can be variously changed within the range which does not deviate from this invention.

  (Supplementary Note 1) A soft decision signal normalization method for normalizing a soft decision signal output from a demodulator, wherein a mode value of a signal amplitude is detected based on a frequency distribution of soft decision signal amplitudes in a decoding unit interval, A soft decision signal normalization method, wherein normalization of the soft decision signal sequence is performed on the basis of the detected mode value.

  (Supplementary note 2) The soft decision signal amplitude consisting of binary numbers is divided into units of powers of 2, and the power value of 2 including the maximum distribution is the most frequent based on the frequency distribution of the soft decision signal amplitude included in each class. The method for normalizing a soft decision signal according to appendix 1, wherein the value is a value.

  (Supplementary Note 3) Demodulation means for soft-decision demodulation of the received signal, mode detection means for detecting the mode value of the signal amplitude based on the frequency distribution of the soft-decision signal amplitude in the decoding unit section of the output of the demodulation means, And normalizing means for normalizing the soft decision signal sequence based on the detected mode value, and decoding means for decoding received data based on the soft decision signal sequence after normalization. Receiving device.

  (Supplementary Note 4) The mode value detection means classifies each soft decision signal amplitude consisting of binary numbers in units of powers of 2, and includes the maximum distribution based on the frequency distribution of the soft decision signal amplitude included in each class. The receiving apparatus according to supplementary note 3, wherein a power value of 2 of a given class is a mode value.

  (Supplementary Note 5) The mode detection means includes a plurality of counters provided for each class in units of powers of 2 and a control means for performing count control of each counter. At each stage, each counter is set to 0 or the maximum value, and the subsequent count output is monitored, and with the input of each soft decision signal, it corresponds to each power of 2 of the class that did not include the soft decision signal amplitude Count up or down only the counters that count, and if the count output of all counters is 0 or no longer the maximum value, all counters are counted down or counted all at once until the count output of any counter reaches 0 or the maximum value Up control is repeated, and the power value of 2 corresponding to the counter whose count output is 0 or the maximum value at the end of the decoding unit interval is set as the mode value. Receiving the additional notes 4, wherein a.

  (Supplementary note 6) The receiving apparatus according to supplementary note 4, wherein the normalizing means divides the soft decision signal amplitude composed of the binary number of the demodulator output by a reference value composed of a power of 2 smaller than the mode value.

  (Supplementary note 7) When N is the number of bits of the soft decision signal output from the demodulator and M is the number of bits of the soft decision signal after normalization, the reference value is limited to a range of 0 to (N−M). The receiving device according to appendix 6, further comprising a restricting unit that performs the above operation.

  (Supplementary Note 8) Demodulation means for soft-decision demodulation of the received signal, mode value detection means for detecting the mode value based on the frequency distribution of the soft-decision signal amplitude in the decoding unit section of the output of the demodulation means, and the detected Normalizing means for normalizing the soft decision signal amplitude based on a reference value based on the mode value, and storing at least a normalized soft decision signal sequence and a reference value related to the initial transmission packet under packet retransmission control Memory, combining means for combining the maximum ratio of each normalized soft decision signal amplitude related to the initial transmission packet and the retransmission packet based on the ratio of the respective reference values, and received data based on the combined soft decision signal sequence A receiving device comprising: decoding means for performing decoding.

(Supplementary Note 9) Demodulating means for soft-decision demodulating a received signal, serial-parallel conversion means for parallel-converting a serial soft-decision signal sequence in a decoding unit section of the output of the demodulating means in units of a predetermined number of symbols, and the parallel Corresponding to each converted soft decision signal sequence, the amplitude of each soft decision signal sequence consisting of binary numbers is divided into units of powers of 2 and the soft decision signal amplitude included in each class A plurality of distribution counting means for counting the frequency distribution of the above, an adding means for adding the count output by class by each distribution counting means for each class, and a class 2 including the maximum distribution based on the frequency distribution of the added output A mode value detecting unit that sets a power value of the mode as a mode value, a normalizing unit that normalizes the soft decision signal amplitude based on the mode value, and a soft decision signal sequence after the normalization. And a decoding means for decoding received data.

It is a principal part block diagram of the receiver by 1st Embodiment. It is operation | movement explanatory drawing of the absolute value conversion and the indexation circuit by embodiment. It is a block diagram of the mode detection circuit by embodiment. It is operation | movement explanatory drawing of the reference value preparation and normalization circuit by embodiment. It is a figure explaining the effect of the normalization system by embodiment. It is a graph of the block error rate by embodiment. It is a principal part block diagram of the receiver by 2nd Embodiment. It is a principal part block diagram of the receiver by 3rd Embodiment. It is a figure which shows an example of a multi-value modulation system. It is a figure which shows the histogram of likelihood distribution in a multi-value modulation system. It is a principal part block diagram of the conventional receiver.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Demodulation circuit 12 Memory 13 Normalization circuit 14 FEC (Forward error Control) decoding circuit 15 Absolute value circuit 16 Indexing circuit 17 Mode value detection circuit 18 Reference value creation circuit 19 Limiting circuit

Claims (8)

A method for normalizing a soft decision signal for normalizing a soft decision signal output from a demodulator, wherein a mode value of a signal amplitude is detected based on a frequency distribution of the soft decision signal amplitude in a decoding unit interval, and the detected mode A soft decision signal normalization method, wherein the soft decision signal sequence is normalized based on a value. The soft decision signal amplitude consisting of binary numbers is divided into units of powers of 2, and the power value of 2 including the maximum distribution based on the frequency distribution of the soft decision signal amplitudes included in each class is set as the mode value. The soft decision signal normalization method according to claim 1. Demodulation means for soft decision demodulation of the received signal;
Mode detection means for detecting the mode value of the signal amplitude based on the frequency distribution of the soft decision signal amplitude in the decoding unit section of the output of the demodulation means;
Normalizing means for normalizing the soft decision signal sequence based on the detected mode value;
A receiving apparatus comprising: decoding means for decoding received data based on the normalized soft decision signal sequence. The mode value detecting means classifies each soft decision signal amplitude consisting of binary numbers in units of powers of 2, and 2 of the class including the maximum distribution based on the frequency distribution of the soft decision signal amplitude included in each class. 4. The receiving apparatus according to claim 3, wherein a power value of is a mode value. The mode detection means includes a plurality of counters provided for each class in units of powers of 2 and control means for performing count control of each counter.
The control means sets each counter to 0 or the maximum value at the initial stage of the decoding unit interval and monitors the subsequent count output, and the soft decision signal amplitude is included with the input of each soft decision signal. If only the counter corresponding to each power of 2 of the class that did not exist is counted up or down, and the count output of all counters is not 0 or the maximum value, the count output of any counter becomes 0 or the maximum value The control to count down or count up all the counters at once is repeated until the counter reaches, and the power value of 2 corresponding to the counter whose count output is 0 or the maximum value at the end of the decoding unit interval is the mode value. The receiving device according to claim 4. 5. The receiving apparatus according to claim 4, wherein the normalizing means divides the soft decision signal amplitude made up of a binary number of the demodulator output by a reference value made up of a power of 2 smaller than the mode value. Limiting means for limiting the value of the reference value within the range of 0 to (N−M), where N is the number of bits of the soft decision signal output from the demodulator and M is the number of bits of the soft decision signal after normalization. The receiving apparatus according to claim 6, further comprising: Demodulation means for soft decision demodulation of the received signal;
A mode value detecting means for detecting a mode value based on a frequency distribution of the soft decision signal amplitude in the decoding unit section of the output of the demodulating means;
Normalizing means for normalizing the soft decision signal amplitude based on a reference value based on the detected mode value;
Under packet retransmission control, a memory that stores at least a normalized soft decision signal sequence and a reference value related to an initial transmission packet;
Combining means for combining the maximum ratio of the soft decision signal amplitudes after normalization related to the initial transmission packet and the retransmission packet based on the ratio of the respective reference values;
A receiving apparatus comprising: decoding means for decoding received data based on the combined soft decision signal sequence.

Images