CN1463084A

CN1463084A - Method and device of iterative demodulation and decode for BPSK modulating system by Turbo encoding

Info

Publication number: CN1463084A
Application number: CN 03137079
Authority: CN
Inventors: 吴晓富; 崔龙; 项海格
Original assignee: COMMUNICATION ENGINEERING COLLEGE SCIENCE & ENGINEEIRNG UNIV PLA
Current assignee: COMMUNICATION ENGINEERING COLLEGE SCIENCE & ENGINEEIRNG UNIV PLA
Priority date: 2003-06-18
Filing date: 2003-06-18
Publication date: 2003-12-24
Anticipated expiration: 2023-06-18
Also published as: CN1209885C

Abstract

The split signal received in the channel is transferred to two sub decoders. With the channel receiving signal being received, the subdecoder 1 calculates out the external information and the information bits size of the equivalent system by using the module in soft-in and sort-out. The sub decoder 2 receives the said information from the output of the decoder 1. Then, the sub decoder 2 calculates and outputs the external information, transferring the information to the decoder 1. Thus, one time of iteration is completed. When the number of iteration reaches a preset value, with the interlacement of the external information output from the decoder 2 being removed, the hard decision is carried out to obtain a estimate value for each information bit, and the decoding procedure is completed. The invention features simple, practical and easy to implement in digital format, suitable to wireless and satellite communications area.

Description

Iterative demodulation decoding method and device for BPSK modulation system of Turbo coding

The technical field is as follows:

the invention relates to the field of communication, in particular to an iterative demodulation decoding method and device for Turbo codes modulated by BPSK in wireless and satellite communication systems.

Background art:

in wireless and satellite communication systems, transmitted bits are subject to channel random noise and random errors are generated. Theory and practice prove that an error correction coding method for providing transmission reliability by introducing redundancy is an effective means. The recently introduced Turbo code is one of the coding schemes with the strongest error correction capability discovered so far.

Turbo codes were proposed by c.berrou equals 1993 [ c.berrou, a.glavieux, and p.thitimajshima, "Near Shannon limit error-correcting and decoding: turbo-codes, "in ICC' 93, Geneva, Swithzerland, May, 1993, pp.1064-1070]It is considered to be one of the largest advances made in coding theory in recent years. Turbo code with error rate of 10 under White Gaussian Noise (Additive White Gaussian Noise-AWGN) channel^-5Time approaches the Shannon (Shannon) capacity limit with a signal-to-noise ratio of 0.7 dB. For satellite channels, because the power is limited, the reduction of the working signal-to-noise ratio of a satellite communication system is undoubtedly of great importance, and the defect of the limited power of the satellite channels can be well solved by introducing the Turbo coding and decoding technology.

Berrou considers only the decoding Algorithm of White Gaussian Noise (AWGN) channel, which is not sufficient in practical wireless and satellite communications. In practical channels, the coded digital signal is generally transmitted by modulation, which requires the same frequency generator at both ends of the transceiver. However, due to the non-ideal characteristics of the actual device, especially due to the time-varying fading characteristics of the transmission medium, the frequency offset and phase of the modulation signal to be transmitted and received are largely unknown in practice, which requires the receiver to perform channel estimation and tracking to eliminate the influence of the unknown frequency offset and phase on the transmitted signal. Therefore, how to remove the influence of unknown channel parameters to decode the Turbo code is a research hotspot which is receiving much attention. The difficulty with this problem is that: turbo codes work with much lower snr than under normal conditions, and in fact how to accurately estimate and track channel parameters at low snr is a big problem in itself.

In satellite and wireless communications, typical channel unknown parameters are frequency offset (frequency offset) and carrier phase. The conventional channel estimation technique uses a phase-locked loop technique. Conventional algorithms generally do not work due to the particularity of communications at very low signal-to-noise ratios (e.g., the signal-to-noise ratio at which Turbo codes can work under AWGN). A possible solution is to give up the separate characteristics of the previous channel estimation and channel decoding and replace them with joint iterative demodulation decoding, so that the channel estimation can take full advantage of the redundancy provided by the encoding.

The method for demodulating and decoding the Turbo code under the unknown phase channel mainly comprises 5 methods by combining the current research dynamics at home and abroad:

method I [ Wuxianfu, Lingcong, Lvjing, iterative demodulation decoding algorithm of Turbo coding under unknown phase channel, electronics declaration, first phase 2002]: the system adopts a scheme of Turbo code plus differential phase modulation, and can improve the performance along with iteration by using an idea of iterative demodulation and decoding based on the proposed optimal incoherent MAP algorithm of differential phase modulation under an unknown phase channel. When the system is under the following conditions: BDPSK modulation is adopted, the number of incoherent MAP states is 32, a Turbo code adopts a 16-state classical structure, the frame length is 2048, and simulation shows that 17 iterations of the system can reach 10 in 3.2dB^-5Hereinafter, the algorithm has better robustness (Robust) to frequency offset. But the defect is that the scheme of adopting differential coding and Turbo code serial cascade partially destroys the structure of the Turbo code, so that the system performance is greatly reduced, and the realization is complex and difficult to realize by the existing FPGA device.

Method two [ g.colavope, "non-coherent Iterative (Turbo) Decoding," IEEE trans. command, vol.48, No.9, sept.2000, pp.1488-1498 ]: iterative detection of the Turbo code incoherent sequence under the unknown phase channel proposed by the Colavolpe. The algorithm does not introduce a differential structure, but directly adopts an incoherent MAP algorithm for received Turbo code signals (BPSK modulation), and compared with the incoherent MAP algorithm which is proposed by us, the algorithm is further simplified, namely a soft incoherent sequence detection algorithm. The algorithm does not damage the structure of the Turbo code, so that the performance is better. But has the disadvantages that the frequency offset effect is not considered and the complexity is still high in implementation.

Method three [ a.antassoulos, and k.m.chung, "Adaptive Iterative Detection for phased tracking in Turbo-Coded Systems," IEEE trans.commun., vol.49, No.12, dec.2001, pp.2135-2144 ]: the Anastasopoulos provides an iterative detection algorithm for phase tracking of a Turbo coding modulation system, and tracks a phase parameter of each path by utilizing a PSP (Per-Survivor-Processing) algorithm, wherein an initial phase is obtained by sending a pilot frequency code. The algorithm is relatively simple to implement and has good performance, but the influence caused by frequency offset is not considered. This algorithm generally does not work well due to the non-linear nature of the frequency offset (the original algorithm was based primarily on the linear nature of the phase).

Method four [ w.oh, and k.cheun, "Joint decoding and carrier phase recovery algorithm for turbo codes," ieee.Commun.Lett., vol.5, No.9, 2001, pp375-377 ]: the joint phase recovery decoding algorithm proposed by wangerok quantizes the channel phase, so that the channel estimation problem is converted into a detection problem, and the detection of the unknown channel phase can be completed by using the principle of the maximum absolute value of soft information (log-likelihood ratio (LLR)) output by Turbo code decoding under the optimal quantization phase. The algorithm is simple to implement and good in performance, and the author also provides a frequency offset estimation algorithm based on interframes. But the tolerated frequency offset is 100ppm of the transmission code rate, and the phase is kept constant in one frame, which obviously hardly meets the requirements of satellite and wireless communication.

Method five [ I.Bar-David, and A.Elia, "Augmented APP (A)²P²)Module for a PosterioriProbability Calculation and Channel Parameter Tracking，”IEEE Commun.Lett.，vol.3，no.1，Jan.，1999，pp.18-20]: Bar-David proposes a demodulation and decoding method of a Turbo coding system under the condition that unknown frequency offset and unknown phase exist in a channel. The structure is that the channel tracking module is organically combined on the basis of the original Turbo code subcode APP module to form the so-called enhanced APP (A)²P²) And a decoding module. The algorithm is simple to implement and good in performance. But the disadvantage is that many key techniques in the original literature are ambiguous and the problem of initial estimation of channel parameters is not well solved. In addition, for the second sub-code, because of the intervention of the interleaver and the deinterleaver, the influence of the channel parameters on the transmitted system information bits is no longer equal to that of the first sub-code (i.e., the system information bits are disturbed by interleaving), and how to calculate the metric of the coded system information bits is to the A of the second sub-code²P²The decoding module is of vital importance to operate, but the literature is silent.

Analysis of the above schemes shows that the problem of Turbo code demodulation and decoding under the condition of fully considering channel frequency offset and unknown phase is not well solved. The frequency offset and realizability (considering existing hardware FPGA, CPLD support) issues are key. The fifth scheme is a preferred choice, which fully considers the tracking problems of frequency offset and phase, and is realizable with the existing FPGA or CPLD device in terms of implementation complexity. The remaining problems are the problem of capturing the initial parameters and how to calculate the metric corresponding to the system information bits, which must be solved by the operation of the second subcode-enhanced APP module, which will be further explained later.

A typical encoder structure of Turbo-code is shown in fig. 1. It is usually composed of two identical Recursive Systematic Convolutional (RSC) codes (usually called subcodes), which RSC1 directly encodes the incoming information sequence to obtain a check sequence y_1k(ii) a At the same time, the information sequence d_kSequence d interleaved by interleaver_nSending the data to RSC2 for encoding to obtain check bit y_2kThe Turbo-code word is composed of information sequence x_kFormed by two paths of check sequences. Check bits (y) generated by the sub-encoder_1k，y_2k) Turbo codes with different code rates can be obtained after the Turbo codes are punctured by different puncturing matrixes.

The Turbo-code iterative decoding structure is shown in fig. 2, and mainly comprises two soft-in soft-out modules (sub-decoders of Turbo codes) for decoding RSC sub-codes in selected Turbo codes. The sub-decoder 1 converts the information bit d obtained by the sub-decoder 2_kExternal information ofAs d_kA priori information is used to decode RSC1 to obtain information about d_kImproved extrinsic information Λ_1e(d_k) Are interwoven to obtainAs a priori information for the sub-decoder 2 to decode RSC 2. The sub-decoder 2 regenerates the extrinsic information a of the information bit improvement in the same manner as the sub-decoder 1_2e(d_j) Is subject to de-interlacing to obtain

As the a priori soft values for the sub-decoder 1 in the next iteration. Thus, after a number of iterations, the output Λ produced by sub-decoder 2₂(d_j) After de-interleaving, hard decision is carried out to obtain each information bit d_kEvaluation of

As can be known from the decoding structure, the key point of decoding the Turbo code in the AWGN channel is the soft-in and soft-out module of the sub-decoder, and in literature, the module has various names, such as BCJR algorithm, MAP algorithm, and APP algorithm. The following adopts the description method of the APP algorithm. A derivation of grid Edge (Edge) based APP algorithms is described in the literature [ Benedetto, "A Soft-Input Soft-Output APP Module for Iterative Decoding of coordinated codes," IEEE Commun. Lett., vol.1, No.1, Jan., 1997, pp.22-24 ].

The documents [ P.Robertson, E.Villebrun, and P.Hoeher, "A compare of optional and sub-optional MAP decoding algorithm in the Log domain," in IEEE int. Conf. on communications (Seatle, WA, June 1995), and pp.1009-1013] show that the MAP algorithm steps can be transferred to the Log domain (so-called Log-MAP algorithm), so that a large number of multiplications in the original MAP algorithm can be replaced by additions and look-up tables, and currently, the actual FPGA implementation commonly adopts the Log domain-based algorithm.

The principle of the APP algorithm based on the Log domain is explained by interpreting Benedetto, and on this basis, the decoding algorithm of the Turbo coding BPSK modulation system under the actual parameter channel is explained.

The RSC sub-code is assumed to be k₀/n₀The binary convolutional code of (1). The following conventions: capital U, C, S, E represent random variables, and lower U, C, S, E represent corresponding implementations each time; using a lower case K to represent discrete time, wherein the time is taken as a discrete time set K; letters I, O represent input and output of the APP module.

Thus, the Log domain based APP module can be described as follows:

1. inputting U ═ U (U)_k)_k∈KRepresenting a sequence of input symbols corresponding to input information bits of a Turbo code subcode encoder (e.g., for an RSC1 subcode, corresponding to d in FIG. 1)_n). For code rate of k₀/n₀Each input symbol U_kFrom k to k₀A bit of information constituting U_k ^j，j＝1，2，Λ，k₀Which realizes u^jE { +1, -1 }. Thus, the Log domain of the prior information distribution of the symbol sequence is denoted as Λ (u; I) ═ Λ_k(u；I))_k∈KWherein

this value is typically derived from the output extrinsic information of another APP module.

2. Output C ═ C_k)_k∈KRepresenting a sequence of output symbols (i.e., code) corresponding to the output information bits of a Turbo code subcode encoder (e.g., for an RSC1 subcode, corresponding to the encoded output (x)_k，y_1k)). For code rate of k₀/n₀Each output symbol C_kFrom n to₀Bit component C_k ^j，j＝1，2，Λ，n₀Which achieves c^jE { +1, -1 }. Thus, the Log domain of the prior probability distribution of the symbol sequence is denoted as Λ (c; I) ═ Λ_k(c；I))_k∈KWherein

this value is a so-called branch metric and is typically calculated from the received samples.

3. Grid description (Trellis Section): as shown in fig. 3, a grid segment from time k to k +1 is taken to illustrate, and state space S ═ S₁，...s_NE.g. state S at time k_kS, S ∈ S), the edges of the mesh segments are given by the product of chi-squared

All possibilities of transition between states at two adjacent times are indicated. In this way, for each edge: initial state s^S(e) End state s^E(e) Input symbols u (e) and output symbols c (e). The specific relationship between these symbols depends on the given recursive convolutional subcode.

4. The purpose of the APP algorithm is to calculate the extrinsic information Λ (u; O) and Λ (c; O) from the input prior information Λ (u; I) and branch metrics Λ (c; I). The state-dependent metrics (Log domain) are first computed backward and forward:

forward metric (FSM):

<math> <mrow> <msub> <mi>A</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>s</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>Σ</mi> <mrow> <msup> <mrow> <mi>e</mi> <mo>:</mo> <mi>s</mi> </mrow> <mi>E</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>s</mi> </mrow> </munder> <mo>&CirclePlus;</mo> <mo>{</mo> <msub> <mi>A</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>S</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>+</mo> <msub> <mi>PR</mi> <mi>k</mi> </msub> <mo>+</mo> <msub> <mi>BM</mi> <mi>k</mi> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </math>

backward metric (RSM):

<math> <mrow> <msub> <mi>B</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>s</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>Σ</mi> <mrow> <mi>e</mi> <mo>:</mo> <msup> <mi>s</mi> <mi>S</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>s</mi> </mrow> </munder> <mo>&CirclePlus;</mo> <mo>{</mo> <msub> <mi>B</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>E</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>+</mo> <msub> <mi>PR</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>+</mo> <msub> <mi>BM</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </math>

wherein,

a metric representing a priori information is provided,

represents Branch Metric (BM) and operates

<math> <mrow> <munderover> <mi>Σ</mi> <mrow> <mi>e</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>N</mi> </munderover> <mo>&CirclePlus;</mo> <mo>{</mo> <msub> <mi>Λ</mi> <mi>e</mi> </msub> <mo>}</mo> <mo>=</mo> <mi>E</mi> <mrow> <mo>(</mo> <msub> <mi>Λ</mi> <mn>1</mn> </msub> <mo>,</mo> <munderover> <mi>Σ</mi> <mrow> <mi>e</mi> <mo>=</mo> <mn>2</mn> </mrow> <mi>N</mi> </munderover> <mo>&CirclePlus;</mo> <mo>{</mo> <msub> <mi>Λ</mi> <mi>e</mi> </msub> <mo>}</mo> <mo>)</mo> </mrow> <mo>,</mo> </mrow> </math>

And E (Λ)₁，Λ₂)＝max(Λ₁，Λ₂)+f_c(|Λ₁-Λ₂I), hereafter we call the E operator, and f_c(x)＝log(1+e^-x) The method can be simply realized by a table look-up method. The initial values are set as:

S₀as an initial state of the convolutional code, S_nIs the final state of the convolutional code. Finally, a backward metric B is calculated_kSimultaneous output of extrinsic information (APP) of(s):

Λ_k(c^j；O)＝LP_k(c^j＝+1；O)-LP_k(c^j＝-1；O) (3)

Λ_k(u^j；O)＝LP_k(u^j＝+1；O)-LP_k(u^j-1; o) (4) wherein,

<math> <mrow> <mrow> <msub> <mi>LP</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <msup> <mi>c</mi> <mi>j</mi> </msup> <mo>;</mo> <mi>O</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>Σ</mi> <mrow> <mi>e</mi> <mo>:</mo> <msubsup> <mi>C</mi> <mi>k</mi> <mi>j</mi> </msubsup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>c</mi> <mi>j</mi> </msup> </mrow> </munder> <mo>&CirclePlus;</mo> <mo>{</mo> <msub> <mi>A</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>S</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>+</mo> <munderover> <mi>Σ</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <msub> <mi>k</mi> <mn>0</mn> </msub> </munderover> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mi>u</mi> <mi>i</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <msub> <mi>Λ</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>u</mi> <mi>i</mi> </msup> <mo>;</mo> <mi>I</mi> <mo>]</mo> <mo>+</mo> <munderover> <munder> <mi>Σ</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> </munder> <mrow> <mi>i</mi> <mo>&NotEqual;</mo> <mi>j</mi> </mrow> <msub> <mi>n</mi> <mn>0</mn> </msub> </munderover> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mi>c</mi> <mi>i</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <msub> <mi>Λ</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>c</mi> <mi>i</mi> </msup> <mo>;</mo> <mi>I</mi> <mo>]</mo> <mo>+</mo> <msub> <mi>B</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>E</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>}</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow> </math>

<math> <mrow> <mrow> <msub> <mi>LP</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <msup> <mi>u</mi> <mi>j</mi> </msup> <mo>;</mo> <mi>O</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>Σ</mi> <mrow> <mi>e</mi> <mo>:</mo> <msubsup> <mi>C</mi> <mi>k</mi> <mi>j</mi> </msubsup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>c</mi> <mi>j</mi> </msup> </mrow> </munder> <mo>&CirclePlus;</mo> <mo>{</mo> <msub> <mi>A</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>S</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>+</mo> <munderover> <munder> <mi>Σ</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> </munder> <mrow> <mi>i</mi> <mo>&NotEqual;</mo> <mi>j</mi> </mrow> <msub> <mi>k</mi> <mn>0</mn> </msub> </munderover> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mi>u</mi> <mi>i</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <msub> <mi>Λ</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>u</mi> <mi>i</mi> </msup> <mo>;</mo> <mi>I</mi> <mo>]</mo> <mo>+</mo> <munderover> <mi>Σ</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <msub> <mi>n</mi> <mn>0</mn> </msub> </munderover> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mi>c</mi> <mi>i</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <msub> <mi>Λ</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>c</mi> <mi>i</mi> </msup> <mo>;</mo> <mi>I</mi> <mo>]</mo> <mo>+</mo> <msub> <mi>B</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>E</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>}</mo> </mrow> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>6</mn> <mo>)</mo> </mrow> </mrow> </math>

as can be seen from (1-6), Λ is required to be determined_k(u^j；O)，Λ_k(c^j(ii) a O), must know Λ_k(u^j；I)，Λ_k(c^j(ii) a I) In that respect Wherein Λ_k(u^j(ii) a I) Actually, the Prior Information (Prior Information) of the Information bits, and the external Information provided by the decoding module of the upper-stage RSC during iteration, what needs to be calculated is the branch metric Λ_k(c^j(ii) a I) This can be provided directly from the received channel samples in an AWGN channel, for example, for a Turbo code with a final rate k formed by a sub-code rate of 1/2 without puncturing the Turbo code₀/n₀(k₀＝1，n₀3), assuming that the system employs BPSK modulation, the coded bits shown in fig. 1 are output (x)_k，y_1k，y_2k) Multiplexing into BPSK Signal stream c_k ^jJ is 1, 2, 3 (corresponding relationship is

c_{k}^{1} = {2 x}_{k} - 1, c_{k}^{2} = {2 y}_{1 k} - 1, c_{k}^{3} = {2 y}_{2 k} - 1),

The effect of the received BPSK signal due to AWGN may be expressed as

<math> <mrow> <msub> <mi>r</mi> <mi>v</mi> </msub> <mo>=</mo> <msubsup> <mi>c</mi> <mi>k</mi> <mi>j</mi> </msubsup> <mo>+</mo> <msub> <mi>w</mi> <mi>v</mi> </msub> <mo>,</mo> <mi>j</mi> <mo>=</mo> <mn>1,2</mn> <mo>,</mo> <mi>Λ</mi> <msub> <mi>n</mi> <mn>0</mn> </msub> <mo>;</mo> <mi>v</mi> <mo>=</mo> <msub> <mi>kn</mi> <mn>0</mn> </msub> <mo>+</mo> <mi>j</mi> <mo>;</mo> <mi>k</mi> <mo>&Element;</mo> <mi>K</mi> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>7</mn> <mo>)</mo> </mrow> </mrow> </math>

Wherein w_vIs a mean of 0 and a variance of σ²White gaussian noise. Is provided with L_c＝2/σ²Then, then

Λ_k(c^j；I)＝L_cr_v，j＝1，2，Λ _n0；v＝kn₀+j；k∈K。 (8)

Now consider the actual channel with unknown parameters, assuming that the system still uses BPSK modulation, and because the frequency offset and phase are unknown, the received signal is matched filtered (sampled once per symbol) and then the optimal sampled complex output is

{\tilde{r}}_{v} = {\tilde{x}}_{v} + i {\tilde{y}}_{v}

(

The number of the I-branch is represented,representing the Q branch) and can be represented as:

<math> <mrow> <msub> <mover> <mi>r</mi> <mo>~</mo> </mover> <mi>v</mi> </msub> <mo>=</mo> <msub> <mi>h</mi> <mi>v</mi> </msub> <msubsup> <mi>c</mi> <mi>k</mi> <mi>j</mi> </msubsup> <mo>+</mo> <msub> <mi>w</mi> <mi>v</mi> </msub> <mo>,</mo> <mi>j</mi> <mo>=</mo> <mn>1,2</mn> <mo>,</mo> <mi>Λ</mi> <msub> <mi>n</mi> <mn>0</mn> </msub> <mo>;</mo> <mi>v</mi> <mo>=</mo> <msub> <mi>kn</mi> <mn>0</mn> </msub> <mo>+</mo> <mi>j</mi> <mo>;</mo> <mi>k</mi> <mo>&Element;</mo> <mi>K</mi> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>9</mn> <mo>)</mo> </mrow> </mrow> </math>

h_v＝exp(jvΩ+Φ₀) Wherein, Ω and Φ₀Respectively representing the channel frequency offset and the initial phase. In general, the channel frequency offset and phase are slowly changing over time (i.e., not only initial estimation but also tracking is needed). Obviously, how to calculate (8) is a problem to be solved, and one approach to solve is to introduce a parameter estimation and tracking mechanism under the framework of APP algorithm.

A proposed by Bar²P²The module well solves the joint parameter tracking and soft-in and soft-out functions of the RSC sub-code under BPSK modulation. Enhanced APP Module (A)²P²) The working principle can be illustrated with fig. 5:

1. channel parameter tracking (CPE) module

The cpe (channel Parameter estimation) module performs a continuous update of the channel parameters by processing the incoming channel samples and outputs to the CSM module. In practical implementation, the CPE is embedded in the APP module, and operates on all Edges (Edges) in the grid graph, the CPE tracks channel parameters on each grid edge and has an operating frequency consistent with the transmission frequency of the coded bits, and the update frequency of each grid edge of the APP is the transmission frequency of the information bits. For code rate of k₀/n₀In the convolutional code, the update frequency of the information outside the APP module is consistent with the update frequency of the grid edge, and is set as a reference working timing sequence k (i.e., a state update timing sequence in the grid), so that the update timing sequence of the CPE in the grid along with the state is also k.For convenience, let us now assume that the value of the channel parameter (frequency offset and phase) estimate at time k state s in the trellis is denoted as χ_k(s)＝(Φ_v(s)，Ω_v(s)). In addition, the CPE updates n on a certain edge₀Next, the process is carried out. The update timing of the channel parameter with respect to the edge is thus v-kn₀+j，j＝1，2，Λ，n₀. In the following description it is assumed that the channel parameters (frequency offset and phase) estimated at time v with respect to a certain edge e (forward or backward) are (for forward use)

For backward use

Representation).

The CPE's workflow is to update n for each edge on the grid segment first₀Next, since multiple edges eventually enter the same final state, the parameters of the final state are based on the metrics on each edge (forward use)For backward use

Means) selecting the best side-estimated channel parameters: (

) As the channel estimation parameter for that state. We now describe in detail.

The CPE is operated according to the following steps:

1) removing (Wipe off operation) the modulation data (i.e. c) corresponding to a certain edge first^v(e)，v＝kn₀+j，j＝1，2，Λ，n₀) Thus obtaining (suppose c)^v(e)∈{+1，-1})

Based on the last tracked parameter estimate phi_v′＝Ω_vT_s+Φ_vIs subjected to phase rotation to obtain

r_v′(e)＝r_v(e)·exp(-iΦ_v′)＝x_v′(e)+y_v' (e), obviously

2) Forward and backward update of channel parameters based on mesh Edges (Edges)

Forward iteration of parameters

Can be written in functional form

Specifically, the following functions are realized:

backward iteration of parameters

The updating of (d) is accomplished by:

wherein,

ΔΦ_v＝aΔΩ_v

a = \frac{2}{3} L T_{s},

l/4, L is a parameter indicating the memory length of the channel parameter estimation.

It should be noted that the initial channel parameters at each trellis edge are determined by the initial state of the edge, that is:

3) for simplicity of implementation, we take:

forward:

wherein,

backward direction:

wherein,

<math> <mrow> <msub> <mover> <mi>m</mi> <mi>ω</mi> </mover> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>B</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>E</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>+</mo> <munderover> <mi>Σ</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <msub> <mi>k</mi> <mn>0</mn> </msub> </munderover> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mi>u</mi> <mi>i</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <msub> <mi>Λ</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>u</mi> <mi>i</mi> </msup> <mo>;</mo> <mi>I</mi> <mo>]</mo> <mo>+</mo> <munderover> <munder> <mi>Σ</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> </munder> <mrow> <mi>i</mi> <mo>&NotEqual;</mo> <mi>j</mi> </mrow> <msub> <mi>n</mi> <mn>0</mn> </msub> </munderover> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mi>c</mi> <mi>i</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <msub> <mi>Λ</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>c</mi> <mi>i</mi> </msup> <mo>;</mo> <mi>I</mi> <mo>]</mo> </mrow> </math>

2. channel sampling parameter modification (CSM) module

The function of the csm (channel Signal modifier) module is to correct the original sampling of the input channel according to the channel parameter value output by the current CPE module, remove the influence of the channel parameter, and thus output a Signal sampling value equivalent to the AWGN channel. In particular, the CSM module is not an independent module, and its function is still embedded in the APP implementation, in fact, compared with the APP algorithm under AWGN, the only difference of the APP algorithm with CSM is that an operation of removing unknown channel parameters must be performed on each edge, and the CPE module functions to obtain updated channel parameters along the edge, so that only corresponding phase rotation is performed on each edge. Is also taken immediately

Re[r_v(e)·exp(-iΦ_v(e))]As input along edges of APP (whereTo take its real part). Others were not as different from standard APP.

3. APP (a) module

The module is basically the same as the APP module of the subcode under AWGN, and the basic function of the moduleIs based on the input Λ_k(u^j；I)，Λ_k(c^j(ii) a I) Calculating Lambda_k(u^j；O)，Λ_k(c^j(ii) a O). The difference lies in the branch metric Λ_k(c^j(ii) a I) And (4) calculating. In detail, this module executes the CPE and CSM modules on the grid graph and on that basis executes equations (1-6), but the branch metrics Λ in (1-6)_k(c^j(ii) a I) Is calculated as follows:

Λ_k(c^j；I)＝L_cRe[r_v(e)·exp(-iΦ_v(e))]，j＝1，2，Λn₀；v＝kn₀+ j; k is e.K (16) where Re [ r ]_v(e)·exp(-iΦ_v(e))]Provided by CSM, CPE module.

Following discussion Turbo code usage A in frequency offset and phase unknown channels²P²The problem with module decoding. A as described above²P²The module can be applied to APP decoding when the RSC subcode adopts BPSK modulation under the condition of unknown frequency offset and phase in principle. However, for the demodulation and decoding of Turbo codes, the following two problems still exist:

1)A²P²the module can be directly applied to decoding Turbo code sub-code RSC1, but the application of sub-code RSC2 requires that the interleaved information symbol sequence is also transmitted to the channel (if the system performance is not deteriorated correspondingly), which is different from AWGN channel because: with respect to the RSC2, the received channel samples contain the portion of the system information bits (i.e., the information symbol sequence x)_kSee fig. 2) no longer corresponds to h due to the effect of interleaving_v＝exp(jvΩ+Φ₀) Because RSC2 sees the relationship at time v as being interleaved, a²P²Necessary for module calculation and corresponding to the first k in RSC code word₀Branch metric a of system information bits_k(c^j；I)，j＝1，2，Λk₀It cannot be calculated. The effectiveness of communication transmission is undoubtedly greatly reduced by additionally transmitting the interleaved information symbol sequence;

2)A²P²the module can only be used for iterative decoding under the condition of parameter tracking, and when the unknown channel frequency offset and phase are larger, simulation shows that A is adopted²P²The algorithm of the module fails.

The invention content is as follows:

the invention aims to provide a method for solving the problem of how to calculate A²P²The iterative demodulation decoding method and device for Turbo code of the measurement problem of the system information symbol sequence of the second sub-code of the module.

Another object of the present invention is to provide a simple and practical iterative demodulation and decoding method and apparatus for Turbo coding BPSK communication system under the channel with unknown parameters.

The invention relates to an iterative demodulation decoding method of a Turbo coding BPSK modulation system, which comprises the following steps:

1) splitting a channel receiving signal and transmitting the split channel receiving signal to two corresponding sub-decoders, namely a sub-decoder 1 and a sub-decoder 2;

2) after receiving the channel receiving signal, the sub-decoder 1 executes a CPE module on the basis of the grid through a soft-in soft-out module thereof to obtain demodulated side estimation channel parameters; executing a CSM module to obtain an equivalent AWGN channel signal; then the obtained equivalent AWGN channel signal is transmitted to an APP module of the soft-in soft-out module, the APP module calculates branch measurement, forward state measurement and backward state measurement, and then output external information and equivalent system information bit measurement are calculated;

3) after receiving the channel signal, the sub-decoder 2 executes a CPE module and a CSM module on the received signal on the basis of a grid through a soft-in soft-out module thereof, transmits the obtained equivalent AWGN channel signal (only corresponding to check bits) to an APP module of the soft-in soft-out module, receives the equivalent AWGN channel signal, the external information output by the sub-decoder 1 and the bit measurement of the interleaved equivalent system information, calculates the output external information, transmits the output external information to the sub-decoder 1, and completes one iteration;

4) the sub-decoder 1 receives the output extrinsic information of the sub-decoder 2, calculates the bit metrics of the output extrinsic information and the equivalent system information again, and the sub-decoder 2 receives the output of the sub-decoder 1, calculates the output extrinsic information again, transmits the output extrinsic information to the sub-decoder 1, and completes the iteration again;

5) after the iteration times reach the preset value, the output external information of the sub-decoder 2 is deinterleaved and then hard decision is carried out to obtain the estimated value of each information bit, and decoding is completed.

The invention carries out initial frequency offset estimation and initial phase estimation on the channel receiving signal and respectively transmits the estimated values to the two sub-decoders.

The iterative demodulation decoding device of the Turbo coding BPSK modulation system comprises two sub-decoders, a sub-decoder 1 and a sub-decoder 2, wherein each sub-decoder comprises a CPE module and calculates demodulated edge estimation channel parameters; a CSM module for calculating equivalent AWGN channel signals; the APP module executes the CPE module and the CSM module on the grid side, receives the outputs of the CPE module and the CSM module, calculates branch metric, forward state metric and backward state metric, and then calculates and outputs external information; the sub-decoder 1 also has an equivalent system information bit metric output end, the sub-decoder 2 has an equivalent system information bit metric input end, and the equivalent system information bit metrics output by the equivalent system information bit metric output end of the sub-decoder 1 are interleaved and input to the equivalent system information bit metric input end of the sub-decoder 2.

The device also comprises an initial frequency offset estimation module and an initial phase estimation module, wherein the initial frequency offset estimation module and the initial phase estimation module are used for respectively calculating an initial estimation value of frequency offset and an initial estimation value of phase and inputting the initial estimation values to the two sub-decoders.

The APP module comprises a BMC module and calculates and outputs branch measurement; the FSMC module receives the output of the BMC module, calculates and outputs forward state measurement; the RSMC module receives the output of the BMC module, calculates and outputs backward state measurement; and the LLR module receives the outputs of the BMC module, the FSMC module and the RSMC module, calculates and outputs the external information and the equivalent system information bit metric for the sub-decoder 1, calculates and outputs the external information for the sub-decoder 2, and calculates and outputs an LLR in the last iteration.

The initial frequency offset estimation module and the initial phase estimation module are solidified on a DSP chip, CPE modules, CSM modules and APP modules of the two sub-decoders are all solidified on a main CPLD chip, and the DSP chip transmits the calculation results of the initial frequency offset estimation module and the initial phase estimation module to the main CPLD chip through an RAM.

The original estimation process of the initial channel parameters of the invention is as follows:

and (3) setting the received baseband complex signal to be given by (9), and comprising the following steps:

1. initial frequency offset estimation

1) First a transformation is made to remove the modulation information:

the above transformation phi is taken as a square, and other transformations are possible, such as phi (x) ═ x | exp (i (2arg (x)), | x | y

Being the modulus of the complex number x, arg (x) denotes the angle at which the complex number x is taken.

2) Performing Fast Fourier Transform (FFT) and searching for frequency offset values

Taking N baseband complex signals, taking 2-power of each complex signal, and finally performing FFT of PN point on the power signals

(PN ═ P × N, P > 1), (P-1) N indicates the number of 0 complements, and thus the frequency offset estimation is obtained by the following equation.

max { | FFT () | } represents the point with maximum amplitude after search FFT transformation, and the sequence number is taken as v_mThen, then

Ω_m＝2πv_m/N。

3) And finally obtaining an initial estimation value of the frequency deviation by using the knowledge of the initial range of the frequency deviation.

Omega calculated by step (2)_mEstimate omega from the final desired initial frequency offset₀There may still be a 180 degree phase jump

Turning to, consider the present frequency offset range Ω₀< 0.1 x 2x pi (this condition is generally satisfied), and finally:

2. initial phase estimation

Conditions are as follows: a small number of pilot symbols are added.

Assume that the system continuously transmits P Pilot Symbols (Pilot Symbols) before the beginning of each frame, which for convenience can be considered to be all 1 codes. The received signal is

The initial phase is estimated as

<math> <mrow> <msub> <mi>Φ</mi> <mn>0</mn> </msub> <mo>=</mo> <mi>angle</mi> <mrow> <mo>(</mo> <munderover> <mi>Σ</mi> <mrow> <mi>v</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>P</mi> </munderover> <msub> <mover> <mi>r</mi> <mo>~</mo> </mover> <mi>v</mi> </msub> <mo>·</mo> <msup> <mi>e</mi> <msub> <mrow> <mo>-</mo> <mi>ivΩ</mi> </mrow> <mn>0</mn> </msub> </msup> <mo>)</mo> </mrow> <mo>.</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>20</mn> <mo>)</mo> </mrow> </mrow> </math>

Therefore, the whole algorithm is simple to realize and can be realized by the current FPGA hardware; particularly in situations where the frequency offset and phase of the channel are unknown. For a typical constant parameter satellite channel, the performance of the Turbo code under an ideal AWGN channel can be obtained.

The iterative demodulation decoding algorithm of the Turbo code under the condition of unknown channel parameters is based on an enhanced APP module, and mainly solves the problems of initial estimation of frequency offset and phase and iterative decoding of a second subcode RSC 2. To solve the iterative decoding of the second subcode RSC2, the present invention proposes an enhanced APP module (hereinafter referred to as ES-A) that provides equivalent system information bit metric output²P²A module).

ES-A²P²Module and A²P²The module principle is basically the same, and the difference is that one more input and one more output are provided. The additional input Λ (s; I) corresponds to an equivalent system information bit metric input. The extra output terminal is equivalent system information bit degreeOutput of quantity, i.e. { Λ_k(s^j；O)，j＝1，2，Λ，k₀(ii) a K is equal to K }, as shown in FIG. 6, the sub-code RSC1 adopts the module to decode and output equivalent system information bit metrics, and after interleaving, the equivalent system information bit metrics are transmitted to an equivalent system information bit metric input end Λ (s; I) of the sub-code RSC2 decoding module, so that ES-A of RSC2²P²The module may use this equivalent information to directly calculate the first k in the codeword corresponding to RSC2₀A of systematic information bits_k(c^j；I)，j＝1，2，Λk₀。

ES-A²P²The module is executed as follows:

1) same as A²P²Modules performing the functions of CPE and CSM modules on a grid basis, see A above²P²Modular principle.

2) APP (a) module

a) Like A²P²And executing (1-6) corresponding modules of the modules. Different points lie in the lambda_k(c^j；I)j＝1，2，Λk₀(ii) a Is calculated if_k(c^j；I)j＝1，2，Λk₀(ii) a Which can be directly calculated by equation (16) from the sampled samples from the channel, the equivalent system information bit metric input Λ (s; I) is discarded (corresponding to the sub-code RSC1 for Turbo codes). Otherwise, executing Λ_k(c^j；I)＝Λ_k(s^j；I)j＝1，2，Λk₀(the sub-code RSC2 corresponding to the Turbo code).

b) Calculating an equivalent system information bit metric output { Lambda ] using the following algorithm_k(s^j；O)，j＝1，2，Λ，k₀；k∈K}：

Λ_k(s^j；O)＝LP_k(s^j＝+1；O)-LP_k(s^j＝-1；O)。 (21)

Wherein

<math> <mrow> <msub> <mi>LP</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <msup> <mi>s</mi> <mi>j</mi> </msup> <mo>=</mo> <msup> <mi>c</mi> <mi>j</mi> </msup> <mo>;</mo> <mi>O</mi> <mo>)</mo> </mrow> <mo>=</mo> <munder> <mi>Σ</mi> <mrow> <mi>e</mi> <mo>:</mo> <msubsup> <mi>C</mi> <mi>k</mi> <mi>j</mi> </msubsup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>=</mo> <msup> <mi>c</mi> <mi>j</mi> </msup> </mrow> </munder> <mo>&CirclePlus;</mo> <mo>{</mo> <msub> <mi>A</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>S</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>+</mo> <mfrac> <mn>1</mn> <mn>2</mn> </mfrac> <msup> <mi>c</mi> <mi>j</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <msub> <mi>Λ</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>c</mi> <mi>j</mi> </msup> <mo>;</mo> <mi>I</mi> <mo>]</mo> <mo>+</mo> <msub> <mi>B</mi> <mi>k</mi> </msub> <mo>[</mo> <msup> <mi>s</mi> <mi>E</mi> </msup> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> <mo>]</mo> <mo>}</mo> <mo>.</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>22</mn> <mo>)</mo> </mrow> </mrow> </math>

The invention mainly solves the problem of iterative demodulation and decoding of the Turbo coding system adopting BPSK modulation under the unknown frequency offset and phase channels. A proposed based on Bar-David²P²Module [ I.Bar-David, and A.Elia, "" Augmented APP (A)²P²)Module for a Posteriori Probability Calculation andChannel Parameter Tracking，″IEEE Commun.Lett.，vol.3，no.1，Jan.，1999，pp.18-20]The invention provides ES-A²P²The module solves the measurement calculation problem of the system information bits required by decoding the second sub code RSC2 of the Turbo code. In addition, the present invention solves the problem of initial parameter estimation. So that ES-A²P²The tracking algorithm of the module can work reliably.

The ES-A provided by the invention²P²Module, compare with A²P²And the module is provided with an additional input end and an additional output end. The method is used for solving the problem of measurement calculation of the system information bit of the second subcode brought by a Turbo code interleaver. Due to the effect of interlacing, A²P²The CSM block of the modules is unable to complete the code word corresponding to RSC2 by effectively removing the effect of the channel parameters (frequency offset and phase)₀Metric Λ of individual system information bits_k(c^j；I)，j＝1，2，Λk₀The computational problem of (2). And ES-A²P²The modules provide additional outputs to compute metrics corresponding to the system information bits, and thus the additional metrics computed by the RSC1 subcode for the system information bits are interleaved and sent to the equivalent system information bit additional inputs of the RSC2 subcode to compute metric problems for the system information bits that are difficult to compute over the channel. Inventive ES-A²P²The module has strong parameter tracking capability, thereby relaxing the requirement on the initial channel parameter estimation precision. Inventive ES-A²P²The module is simple to realize and can be realized by the existing FPGA technical hardware.

The invention provides a channel parameter initial estimation algorithm which can be effectively used for the problems of frequency deviation and phase estimation under an extremely low signal-to-noise ratio. For frequency offset estimation, the method does not need pilot frequency symbols, improves the utilization rate of a channel, and overcomes the defect that the common frequency offset estimation is difficult to work under low signal-to-noise ratio (lower than 0 dB). A small number of pilot frequency symbols are utilized, and an initial estimation algorithm of a channel phase is provided on the basis of frequency offset estimation. The estimation accuracy of the parameters can generally ensure the ES-A of the invention²P²The module can operate reliably on typical satellite channels.

The advantages and positive effects of the invention are summarized as follows:

1. the invention provides an ES-A (extended sequence-adaptive) adaptive to iterative demodulation and decoding of A Turbo coding BPSK (binary phase shift keying) system under frequency offset and phase unknown channels²P²Module, compare with A²P²And the module is provided with two additional input and output ends. Can be used to effectively solve the problem of the second sub-code brought by the Turbo code interleaverMetric computation problem of system information bits. (ii) a

2. ES-A of the present invention²P²The module is simple and practical when aiming at the special input conditions in the step 1 and is easy to realize digitally;

3. the initial frequency offset estimation algorithm can effectively work under the condition of low signal-to-noise ratio (lower than 0dB), and is easy to rapidly realize by adopting a digital signal processing technology;

4. the initial phase estimation algorithm of the invention adopts A small amount of pilot frequency symbols on the basis of initial frequency offset estimation, the realization is simple, and the estimation precision can meet the requirement of ES-A²P²The need for module tracking;

description of the drawings:

FIG. 1Turbo-code encoder architecture;

FIG. 2Turbo-code iterative decoder architecture;

FIG. 3 is a trellis coded diagram;

edges of the grid of FIG. 4 (An Edge of the trellis section);

FIG. 5 enhanced APP Module (A)²P²)；

Wherein: CPE: a channel parameter estimation module; CSM: a channel parameter removing module for the channel input signal;

APP (a): a soft-in soft-out module of the subcodes; r is_kBaseband complex signal representing input

FIG. 6 is a block diagram of an iterative decoding algorithm of a Turbo code BPSK system under unknown channel frequency offset and phase according to the present invention;

FPGA development structure schematic diagram of FIG. 7 example

Din denotes raw data, Ain denotes a priori information, Aout denotes extrinsic information, and Px0 denotes equivalent bit output. If ES-A²P²The module needs to average the estimated phases of FSMC and RSMC, then the portion shown in dashed box is added. In actual implementation, FSMC and BMC are calculated together, and RSMC and BMC are the same.

FIG. 8 illustrates an RSC encoder used in the example;

FIG. 9 is a schematic block diagram of a decoding hardware implementation;

FIG. 10 is a FSMC module implementation cell;

FIG. 11 a BMC module implementation unit;

FIG. 12LLR module unit;

FIG. 13 is a trellis diagram of the RSC used in the example;

FIG. 14 is a system performance diagram of an example

The abscissa of the graph is the signal-to-noise ratio Eb/N0, the ordinate is the Bit Error Rate (BER), and 4 curves in the graph represent the performance after the 1 st to 4 th iterative decoding from top to bottom respectively

The implementation scheme is as follows:

the overall scheme of the iterative demodulation decoding method of the present invention is shown in fig. 6.

The embodiment provides a development example of a Turbo code demodulation and decoding algorithm which is proposed in the third generation mobile communication system IMT-2000 high-speed service and is based on a CPLD chip EP20K400EBC652-2X of the Alter company under a typical frequency offset and phase unknown channel. The FPGA implementation of this example is shown in fig. 7, implementing a master clock of 32MHz, allowing a decoding rate of 115.2 kbps. The system has 5 additional external RAM blocks and one DSP chip (TMSC5402) where initial frequency offset and phase estimation is done. The division operation in the CPE module is realized by using a table lookup. Before further explanation of fig. 7, we first give specific coding parameters for the Turbo code chosen in the example.

The embodiment selects the channel coding of the third generation mobile communication system IMT-2000 high-rate service.The Turbo code of the preferred proposal consists of two identical recursive convolutional subcodes whose generator polynomials are (13, 15, 17)₈Information frame length 2280, interleaving length 2298, total length 2304, and coding rate 1/2. The 1/2 rate Turbo code is obtained by concatenating two RSC mother codes and deleting it, as suggested by IMT 2000. The mother code selected for this example is shown in fig. 8.

The code rate adjustment in the Turbo code encoder is to adjust the code rate by deleting some check bits. In table 1, "1" indicates output, and "0" indicates deletion; (X, Y)₀，Y₁) Representing the output of a first RSC encoder, where X is an information bit and Y is₀And Y₁Is check digit, (X ', Y'₀，Y’₁) Representing the output of the second RSC encoder, X 'being information bit, Y'₀And Y'₁Is a check bit.

Note that for the 1/2 rate Turbo code, the Y1 and Y' 1 bit spaces are not used, and the scope of erasures is not included in the following description of this example. The erasure indication p is used only to tailor the erasure case for the Y0 and Y' 0 bits (p-0 means the bit is erased and p-1 means the bit is not erased).

	Code rate R
	Code rate R			Output of	1/2	1/3	1/4
X	11	11	11	Output of	1/2	1/3	1/4
X	11	11	11	Y₀	10	11	11
Y₁	00	00	10	Y₀	10	11	11
Y₁	00	00	10	X’	00	00	00
Y’₀	01	11	01	X’	00	00	00
Y’₀	01	11	01	Y’₁	00	00	11

Code rate adjustment for non-return-to-zero bits in a table

Note: for each code rate, the table is read from top to bottom and from left to right

Assuming in this example that the channel is a typical satellite constant reference channel, the initial normalized frequency offset (relative to the symbol transmission rate) of the channel is assumed to be randomly chosen at Δ fT ∈ (0, 0.1), while the unknown phase of the channel is the phase between the randomly chosen (0, 2 π).

Selecting algorithm parameters:

(1) to estimate the initial phase, the number of pilots transmitted is 20 per frame.

(2) Initial frequency offset estimation (22-23) the algorithm, the FFT length is selected to be N1024, and P is 1;

(3) the iterative decoding of the system adopts the algorithm shown in fig. 5, wherein the CPE module parameters are selected as: channel with a plurality of channels

Memory length L128;

the overall framework of demodulation decoding for the above scenario is shown in fig. 6.

FIG. 9 is A schematic block diagram of an iterative decoding algorithm hardware implementation of A CPLD for implementing A Turbo code under unknown frequency offset and phase, in order to save on-chip resources, and according to the practical situation of the algorithm-two ES-A²P²The sub-decoders can not work simultaneously, and the same ES-A is time division multiplexed when the hardware is realized²P²And (5) modules.

The key in this decoder is ES-A²P²The CPLD implementation structure of the sub-decoder is shown in FIG. 7. The main chip consists of a piece of EP20K400EBC652-2X of Altera company, a piece of DSP chip TMSC5402 of TI company and 6 external RAM chips. The DSP chip is used for estimating initial frequency deviation and carrier phase, works only at the beginning of each frame, and transmits the result to the main CPLD chip through the RAM to start ES-A²P²And a core decoding algorithm module.

And ES-A²P²There are four main parts in the algorithm module: forward state metric calculation module (FSMC), backward state metric calculation module (RSMC), log likelihood ratio calculationA module (LLR) and a branch metric calculation module (BMC). And 6 pieces of external RAM (A, B, C, D, E, F) are used to store intermediate calculation results for the FSMC module and the RSMC module. Where an external RAM-F is used to store frame data (received samples) from the channel, the initial parameter estimates computed by the DSP chip are also stored on the chip RAM.

The computing structures of the specific FSMC and the RSMC are very similar, and the computing process is described by taking the FSMC as an example. For FSMC, there are two options: parallel computation and serial computation. If parallel computing is adopted, the clock needed by computing an FSM is less, but the hardware resource needed by computing is more than that of serial computing, and the computed data needs to be stored in an off-chip RAM at the same time and read at the same time, so that the requirements on the size of the off-chip RAM and the hardware control are higher. We use serial calculations (as in fig. 10).

As can be seen from FIG. 10, the implementation flow of FSMC is not much different from the implementation flow of corresponding FSMC in the general APP decoding algorithm, and ES-A is actually used²P²The difference from APP is mainly reflected in BMC. ES-A²P²In BMC (as shown in FIG. 11), the influence of frequency offset and phase offset is removed, so that ES-A²P²The FSMC module in (1) is the same as the FSMC in the common APP.

We perform LLR calculation serially as well, as in fig. 12. In RSC1, we need to compute the equivalent output bit metrics (PS0) and the extrinsic information is supplied to RSC2 for use, in RSC2 we need only compute the extrinsic information if it is not the last iteration; if it is the last iteration, we only need to calculate the final log-likelihood ratio LLR.

And grid-based ES-A²P²The execution timing of the modules is shown in fig. 10. The method comprises the following specific steps:

1. calculating initial frequency offset omega of channel by using formula (22-25) for each frame_cAnd initial phase phi_cThe specific implementation can be implemented on a Digital Signal Processing (DSP) chip, and the calculated parameter is set to χ_c＝(Φ_c，Ω_c). This parameter is sent to ES-A of RSC1²P²A module;

2. the decoder performs iterative decoding based on the framework of fig. 6 (and based on fig. 9 in hardware implementation), and in one iteration, the RSC1 subcode is decoded first, and then the RSC2 subcode is decoded, so that the next iterative decoding is continued after one iteration is completed;

3. when the iteration time t is 1, the ES-A of RSC1 subcode is executed²P²And (5) modules. The ES-A²P²The modules proceed based on the trellis timing as in fig. 13, and for an RSC subcode with code rate 1/2, trellis time k is 0, 1, 2. And code words (X, Y) on the edges (Edge) between two adjacent states₀，Y₁) Has a bit timing of 2k + j, j being 1, 2.

1) At grid time k 0, ES-A²P²The module performs the following initialization tasks: performing parameter assignment χ on all states₀(s₀)＝χ_c，s₀＝0，1，2，...，7。

2) At the grid time k is 0, the forward iteration metric A is taken_kThe initial values of(s) are set to:

S₀is the initial state of the convolutional code;

3) from the grid time k-1 to the grid time k, based on the Edge (Edge), from A_k-1(s) calculation of A_k(s). As in fig. 13, for a certain state s at the time k of the trellis_kIt is assumed that it can be represented by two states s at trellis time k-1_k-1 ¹And s_k-2 ²Transfer to obtain the corresponding edge as e₁And e₂(ii) a Opposite edge e₁；

(a) First setting an initial value

(b) CPE module works based on bit timing of code word on edge, calculates in succession

Wherein p ═ 1 represents the compound

The corresponding bit of the RSC subcode is not deleted, and p is 0 to indicate that the bit is deleted;

(c) calculating the forward metric for each edge according to equations (14) (16)Find out

Then updates the corresponding state s at the grid time k_kIs/are as follows

Parameter(s)

(d) While calculating a forward metric A according to equations (1) (16)_k(s)。

4) At grid time k 2298, the forward iteration ends and the backward iteration begins. Backward iterative metric B_kInitial value setting of(s):S_nis the final state of the convolutional code;

5) from grid time k +1 to grid time k, based on Edge (Edge), from B_k+1(s) calculation of B_k(s); as in fig. 13, for a certain state s at the time k of the trellis_kSuppose it can be represented by two states s at trellis time k +1_k+1 ¹And s_k+2 ²Transfer to obtain the corresponding edge as e₁And e₂(ii) a Opposite edge e₁。

(e) First setting an initial value

(f) CPE module works based on bit timing of code word on edge, calculates in succession

Wherein p ═ 1 represents

(g) calculating a backward measure for each edge according to equations (15) (16)

Find out

Then updates the corresponding state s at the grid time k_kIs/are as follows

Parameter(s)

(h) Calculating the backward measure B according to the formulas (2) and (16)_k(s)；

(i) Calculating Lambda according to equations (4), (6) and (16)_k(u^j(ii) a O) and output, calculating Lambda from (21), (22) and (16)_k(s^j；O)

And outputting;

6) and (5) the backward iteration is finished until the grid time k is equal to 0.

4. When the iteration time t is 1, the ES-A of RSC2 subcode is executed²P²And (5) modules. The ES-A²P²The module proceeds based on the grid timing as in fig. 13, except with respect to Λ_k(c^j(ii) a I) Calculation of j-1, which implements ES-A completely analogous to RSC1 described in step 3²P²Module, and Λ_k(c^j(ii) a I) The calculation of j-1 cannot be performed by (16) since the channel does not provide a corresponding sample of the system information bits of RSC1, but instead the Λ output by the RSC1 subcode is changed here_k(s^j(ii) a O) providing, the output is set to be Lambda after interleaving_k(s^j(ii) a I) Put Λ_k(c^j；I)＝A_k(s^j(ii) a I) J is 1, the calculation can be completed;

5. if the iteration time t is equal to the preset iteration time, outputting a decoding value, otherwise, setting t to t +1, and going to 3 to continue execution.

The performance of this example over 4 iterations under a typical constant reference channel is shown in fig. 14, from which fig. 14 can be seen:

1. BPSK modulation is adopted, under 4 iterations, the performance difference between the text algorithm under unknown frequency offset and phase and the performance under AWGN is only within 0.4 dB;

2. the algorithm complexity is as follows: the algorithm complexity of the invention is about 2 times of that of the decoding algorithm under AWGN.

While the invention has been shown and described with reference to certain particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. An iterative demodulation decoding method of a Turbo coding BPSK modulation system comprises the following steps:

2) after receiving the channel receiving signal, the sub-decoder 1 executes a CPE module on the basis of a grid through a soft-in soft-out module thereof to obtain demodulated side estimation channel parameters, executes a CSM module to obtain an equivalent AWGN channel signal, then transmits the obtained equivalent AWGN channel signal to an APP module of the soft-in soft-out module, and the APP module calculates branch metric, forward state metric and backward state metric, and then calculates output external information and equivalent system information bit metric;

3) after receiving the channel signal, the sub-decoder 2 executes a CPE module and a CSM module on the received signal on the basis of a grid through a soft-in soft-out module thereof, transmits the obtained equivalent AWGN channel signal to an APP module of the soft-in soft-out module, receives the equivalent AWGN channel signal, the external information output by the sub-decoder 1 and the bit measurement of the interleaved equivalent system information, calculates and outputs the external information, and transmits the external information to the sub-decoder 1 to finish one iteration;

2. The iterative demodulation and decoding method for a Turbo encoded BPSK modulation system as claimed in claim 1, wherein an initial frequency offset estimation and an initial phase estimation are performed on the channel received signal, and the estimated values are transmitted to the two sub-decoders, respectively.

3. An iterative demodulation decoding device of a Turbo coding BPSK modulation system comprises two sub-decoders, a sub-decoder 1 and a sub-decoder 2, wherein each sub-decoder comprises a CPE module and calculates demodulated edge estimation channel parameters; a CSM module for calculating equivalent AWGN channel signals; and an APP module, executing the CPE module and the CSM module on the grid edge, receiving the output of the CSM module of the CPE module, calculating branch metric, forward state metric and backward state metric, and then calculating output external information; the method is characterized in that the sub-decoder 1 is also provided with an equivalent information bit output end, the sub-decoder 2 is provided with an equivalent information bit input end, and equivalent information bits output by the equivalent information bit output end of the sub-decoder 1 are input to the equivalent information bit input end of the sub-decoder 2 after being interleaved.

4. The iterative demodulation and decoding apparatus for a Turbo encoded BPSK modulation system as claimed in claim 3, further comprising an initial frequency offset estimation module and an initial phase estimation module for calculating an initial estimation value of frequency offset and an initial estimation value of phase, respectively, and inputting the initial estimation values to the two sub-decoders.

5. The iterative demodulation and decoding apparatus for Turbo encoded BPSK modulation system of claim 3 or 4, wherein said APP module comprises a BMC module, which calculates and outputs branch metrics; the FSMC module receives the output of the BMC module, calculates and outputs forward state measurement; the RSMC module receives the output of the BMC module, calculates and outputs backward state measurement; and the LLR module receives the outputs of the BMC module, the FSMC module and the RSMC module, calculates and outputs the external information and the equivalent information bit for the sub-decoder 1, calculates and outputs the external information for the sub-decoder 2, and calculates and outputs an LLR in the last iteration.

6. The iterative demodulation and decoding device of Turbo coding BPSK modulation system according to claim 5, wherein the initial frequency offset estimation module and the initial phase estimation module are fixed on a DSP chip, the CPE module, the CSM module and the APP module of the two sub-decoders are fixed on a main CPLD chip, the DSP chip transmits the calculation results of the initial frequency offset estimation module and the initial phase estimation module to the main CPLD chip through RAM; the intermediate calculation results of the respective modules are stored in the RAM.

7. The iterative demodulation and decoding apparatus for a Turbo encoded BPSK modulation system as claimed in claim 5, wherein the sub-decoder 1 and the sub-decoder 2 are the same decoder time-division multiplexed.