WO2004047355A1

WO2004047355A1 - Method and apparatus for channel coding and rate matching

Info

Publication number: WO2004047355A1
Application number: PCT/EP2003/011897
Authority: WO
Inventors: Wen Xu
Original assignee: Siemens Aktiengesellschaft
Priority date: 2002-11-15
Filing date: 2003-10-23
Publication date: 2004-06-03

Abstract

Method for channel coding (102) and rate matching (103) of information bits (12, 13) includes the steps of i) encoding (102) first data bits (12) to code bits (13) using a recursive systematic convolutional channel encoding method, and ii) rate matching (103) said code bits (13) by repeating or/and puncturing the code bits (13).

Description

Title

METHOD AND APPARATUS FOR CHANNEL CODING AND RATE MATCHING

Description

Field of the Invention

Generally, the invention relates to channel coding and rate matching. More specifically, the invention relates to aforementioned operations in a mobile communications environment, such as in mobile telecommunication network, especially at GSM/EDGE base stations and terminals.

Background of the Invention

The worldwide most successful mobile communication system GSM is further evolving towards the third generation within the framework of 3GPP GERAN (3^rd generation partnership project GSM/EDGE Radio Access Network) (at the time of writing, November 15, 2002, see http://www.3gpp.org). Goal of the

GERAN evolution is to realize similar services provided by Universal Mobile Telephony System (UMTS) Radio Access Network (RAN) UTRAN.

In the GERAN services are basically offered through optimised transport channels, and generic transport channels. The optimised channels are set up based on the exact knowledge of the service, so that an efficient transmission is possible. An example is the GSM speech transmission, where the speech- encoded bits are channel-coded through Unequal Error Protection UEP. The generic channels, on the other hand, are established without exact knowledge of the service, and correspondingly the Equal Error Protection EEP is employed. Examples are (E)GPRS and ECSD. Optimized channels offer a good spectral efficiency, but require a specific channel coding scheme for each service. In contrast, the generic channels do not require individual channel coding schemes, but the EEP delivers in many cases a bad performance. For instance, a speech transmission through an EGPRS channel may lead to a performance loss of about 5 dB, compared with the optimized channel.

Especially due to bad performance of real time packed switched (such as the Internet Protocol IP) speech and multimedia services based on the existing (E)GPRS radio bearers, new physical layer concepts such as the Flexible Layer One FLO concept based on the UTRAN physical layer (layer one) scheme are under current standardization discussions within 3GPP GERAN. Initial investigations showed significant performance improvements by using the FLO, in comparison with the (E)GPRS based radio bearers.

Most systems of communications are designed according to the Open Systems Interconnection OSI reference model which defines 7 layers describing all functions of a network from the physical medium up to the application. The first layer or layer one, on which all others rely, is the one which accomplishes the transmission, sometimes referred as the physical layer. The problem which has to be solved is that a 1 is received whenever a 1 is sent, and a 0 is received whenever a 0 is sent. The main tasks of the layer one include channel coding/decoding and modulation/demodulation, although it- is also in charge of other tasks, such as power control, synchronization, hand-over monitoring, etc. With the Flexible Layer One, the physical layer of GERAN offers several Transport Channels TrCH to the layer 2, the so-called Medium Access Control MAC sublayer. Each of these transport channels can carry one data flow providing a certain Quality of Service QoS . A number of TrCHs can be multiplexed and sent on the same physical channel at the same time.

The configuration of a TrCH, i.e., the number of input bits, parity check is denoted as the transport format. Like in UTRAN, a number of different transport formats can be associated to one TrCH. The configuration of the transport formats is controlled by the Radio Access Network and signaled to the mobile station MS at call setup. In both the mobile station and the base station BS, the transport formats are used to configure the encoder and decoder units .

On each TrCH, transport blocks are exchanged between the MAC sublayer and the physical layer on a transport time interval basis of 20ms. For each transport block (i.e., for each TrCH, every 20ms) a transport format is chosen and indicated through the Transport Format Interval TFI. Only a limited number of combinations of the transport formats of the different TrCHs are allowed. In order to decode the received sequence the receiver needs to know the Transport Format Combination for a radio block. This information is transmitted in the Transport Format Combination Indicator TFCI field. This Layer One header is firstly decoded by the receiver. From the decoded TFCI value, the transport formats for the different transport channels are known and the actual decoding can start. Figure 1 shows the basic principle of the Flexible Layer One

FLO concept already discussed in the standardisation forum.

The Rate Matching step 103 is the core of the FLO concept. It punctures or repeats the bits of an Encoded Block 13 so that a given number of output bits can be generated for a given number of input bits. Since the number of bits on a transport channel can vary between different transmission times, bits are repeated or punctured to ensure that the total bit rate after TrCH multiplexing is identical to the total bit rate available on the physical channel.

A FLO channel coder and rate matcher apparatus 10 operates in the following way: Transport Block 11, corresponding to a Traffic Channel TrCH(i) is passed to step 101 from OSI Layer 2. The channel coding and matched rate matching apparatus 10 operates the Transport Block 11 to Radio Block 16, after which the Radio Block 16 is transmitted over a Physical Channel .

In step 101 the FLO channel coding and rate matching apparatus 10 attaches the Cyclic Redundancy Check CRC bits to the Transport Block 11, thus producing a Code Block 12. The Code Block 12 is transformed to Encoded Block 13 in the Channel Coding step 102. The Encoded Block 13 is then transformed to a Radio Frame 14 in the Rate Matching step 103. The Radio Frame 14 is in step 104 multiplexed for transport channels, the multiplexed frame being a CCTrCH frame 15. For the CCTrCH frame 15 is then performed TFCI Mapping (step 105) and interleaving (step 106) , after which the CCTrCH frame 15 has been tranformed to a Radio Block 16. Similar oparation is performed to each block TrCH(i+l), TrCH (i+2) , ... so long as the data transfer is ongoing. The effective coding rate (i.e., coding rate after puncturing or repetition = number of information bits / number of bits after rate matching) for each TrCH depends on the Transport

Block 11 sizes and associated rate matching attributes for all TrCH. This attribute is an integer between 1 and 256, usually coded with 8 bits, defining priorities between the different transport channels. The higher the rate matching attribute is, the lower the coding rate is.

For every Radio Block 16 to be transmitted, one radio frame from each active TrCH is delivered to the TrCH multiplexing. These radio frames are serially multiplexed into a Coded Composite Transport Channel CCTrCH.

The rate matching algorithm which is foreseen for GERAN is based on the generic uniform rate matching algorithm specified in [5] . A few simplifications need to be made since there is neither spreading factor, nor compressed mode, and therefore many parameters of the UTRAN algorithm can be fixed either to 0 or 1 in GERAN.

Several convolutional codes for the application in Flexible Layer One FLO were investigated in [1] and [2] . It has showed that using the code with generator polynomials G4, G5, G7, which is in fact used already in EGPRS, sufficient performance can be achieved [1] . The previous investigations have utilised Non-Recursive Non-Systematic Convolutional NRNSC codes.

Although the overall performance of such solutions has been pretty good, it is still far from being satisfactory. Therefore, it remains a problem to further develope the suitable methodology to be used in apparata, in other words, to find suitable convolutional codes with which better performance can be obtained.

It is an object of the invention to obtain a solution by means of which it is possible to bring about an apparatus and a method for more efficient channel coding and data rate matching.

Brief Description of the Invention This and other objectives of the invention are accomplished in accordance with the principles of the present invention, in the manner described in the patent claims, by providing an FLO concept based on a Recursive Systematic Convolutional RSC code and a corresponding generic Rate Matching RM algorithm matched to the RSC code. As a concrete example, the RSC code is the one with generator polynomials [G4/G4, G7/G4, G5/G4] , and the RM algorithm can be the generic RM algorithm standardized for Turbo encoded traffic channels in UTRAN [5] .

In comparison with the FLO realized with an NRNSC code and appropriate rate matching algorithms, an RSC coder with the same code rate (and even the same polynomials used) offers significant performance improvements in terms of Bit Error Rate BER for effective coding rates greater than 1/3, the rate of the mother code. For other coding rates there is no degradation. Furthermore, the RSC coder offers inherently a *no coding" option without the need for additional signaling.

An apparatus for flexible layer one coding and rate matching, includes i) convolution means, especially corresponding a recursive systematic convolutional channel encoder, adapted to encode information bits, ii) code generator for generating code bits from said encoded information bits, and iii) rate matching means, preferably adapted to utilise a generic rate matching algorithm, adapted to repeat or/and puncture the code bits.

According to one aspect of the present invention, the coding and/or decoding method and/or apparatus is used in a mobile terminal, such as a mobile phone. An advantage of such a solution is that it offers more reliable data transmission, with no more computational power and the design complexity in mobile terminals. For the same QoS, the present method can further increase data rates in transmission.

According to one aspect of the present invention, the coding and/or decoding method and/or apparatus is used in a base station of a mobile network. A advantage of such a solution is that it offers more reliable data transmission, with no more computational power and the design complexity. For the same QoS, the present method can further increase data rates in transmission, or for the same data rate, more customers can be served.

Brief Description of the Drawings In the following, the invention and its preferred embodiments are described more closely referring to the examples shown in Figures 2-7 in the appended drawings, wherein:

Figure 1 presents the structure of the FLO concept; Figure 2 shows simulation results for effective coding rate = 1;

Figure 3 shows simulation results for effective coding rate = 0,75;

Figure 4 shows simulation results for effective coding rate = 0.44;

Figure 5 shows simulation results for effective coding rate = 1/3;

Figure 6 shows simulation results for effective coding rate = 1/6;

Figure 7 illustrates a system wherein the apparatus and method can be used; and

Figure 8 is a block diagram of the apparatus.

Detailed description of the invention

Interest in Recursive Systematic Convolutional RSC codes was inspired by the performance advantages compared to Non- Recursive Non-Systematic Convolutional NRNSC codes (see e.g. [4] for the adaptive mutirate speech codec standard) .

Furthermore, RSC codes with subsequent puncturing offer a no coding" option without the need for additional signaling.

If systematic codes are to be used, the rate matching algorithm has to be changed since puncturing of systematic bits would degrade the performance significantly and pose the risk to render the code catastrophic. An equivalent problem has been solved already for UTRAN, which results in the RM algorithm for Turbo encoded traffic channels, or simply, for

Turbo code [5] .

Recursive systematic convolutional (RSC) encoder

For the RSC encoder, the same polynomials as proposed for the NRNSC encoder were chosen (G4, G7, G5) with G4 as feedback polynomial, G4/G4 = 1

G7/G4 = 1 + D + D² + D³ + D⁶/ 1 + D² + D³ + D⁵ + D⁶

G5/G4 = 1 + D + D⁴ + D⁶ / 1 + D² + D³ + D⁵ + D⁶

For the information bit sequence {u(k)} (k = 0, 1, ..., L-l, L is the size of the information bit sequence) , the code bit sequence {C(3k), C(3k+1), C(3k+2)} can be calculated as r(k) = u(k) + r(k-2) + r(k-3) + r(k-5) + r(k-6)

C(3k) = u(k)

C(3k+1) = r(k) + r(k-l) + r(k-2) + r(k-3) ^■»- r (k-6)

C(3k+2) = r(k) + r(k-l) + r(k-4) + r(k-6)

And for termination of the coder, the tail code bits can be calculated as r(k) = 0

C(3k) = r(k-2) + r(k-3) + r(k-5) + r(k-6)

C(3k+1) = r(k) + r(k-l) + r(k-2) + r(k-3) + r(k-6)

C(3k+2) = r(k) + r(k-l) + r(k-4) + r(k-6)

Rate matching algorithm for RSC code

In contrary to NRNSC codes where all bits are equally eligible for puncturing, a rate matching algorithm for systematic codes should avoid puncturing of systematic bits

(here, the code bits C(3k), corresponding to the polynomial

G4/G4) . This can be achieved by the following alternatives:

1) Using the generic RM algorithm for Turbo code:

The RM algorithm for Turbo encoded traffic channels deals with 3 bit sequences, namely systematic bit sequence, parity bit sequence generated by the first constituent encoder, and parity bit sequence generated by the second constituent encoder. The whole description of this rate matching algorithm can be found in Section 4.2.7 of [5] .

The code bits sequence C(0), C(l), C(2), C(3), ..., C(3k), ..., C(3k), C(3k+1), C(3k+2), ... can be simply used as input to the RM algorithm for Turbo code, with C(3k) as the systematic bit sequence, C(3k+1) corresponding to the first parity bit sequence and the C(3k+2) the second parity bit sequence. Within the RM algorithm for Turbo code, and in the case of puncturing, the systematic code bits are not punctured. The first parity bit sequence and the second parity bit sequence are treated separately, i.e. by using the generic uniform RM algorithm for convolutional code (see below) twice, with different initialisation parameter sets (such as starting offsets) .

2) Using the generic uniform RM algorithm: The generic uniform RM algorithm is in fact the basic RM algorithm standardized in UTRAN (see below for description) , originally designed for NRNSC code bits. It repeats or punctures the input bits as uniformly as possible, without taking into account the properties of the channel code (such as the free distance, distance spectra, ...) . We can employ the generic uniform RM algorithm as follows.

In the case of repetition, all the code bit sequences (C(3k),

C(3k+1), C(3k+2)} are fed to the generic RM algorithm. I.e. there is no difference between RSC and NRNSC code bits in repetition mode.

In the case of puncturing, the encoded bit sequence (= the output of the RSC coder) is divided into sequences containing the systematic bits C(3k), and the parity bits C(3k+1) and C(3k+2). Systematic bits C(3k) will not be punctured and sent to the output of the RM algorithm directly. The parity bits C(3k+1) and C(3k+2) are properly collected together, just like the normal code bits of an NSNRC encoder, and sent to the generic uniform RM algorithm for puncturing.

In the above alternatives, some systematic tail code bits C(3k) (generated by G4/G4) need to be considered as parity bits and punctured as well, when the number of the bits to be punctured is greater than the number of all the normal parity bits C(3k+1) and C(3k+2). This may happen when the effective code rate is close to 1. Concretely, we can assign the systematic tail code bits to the parity bit sequence to be fed to the rate matching algorithm for puncturing.

We can further optimize the whole performance of channel coding and rate matching by taking into account the properties of the channel code, or the importance of the channel code bits. In our example, the systematic bits are much more important than the parity bits, i.e. the generator polynomial G4/G4 (= 1) is stronger than the generator polynomial G7/G4 or G5/G4. This means that puncturing systematic bits generated by G4/G4 usually leads to more performance loss than puncturing the same amount of parity bits generated by G7/G4 or by G5/G4, or repeating systematic bits generated by G4/G4 usually leads to more performance gain than repeating the same amount of parity bits generated by G7/G4 or by G5/G4.

Comparing different parity bits, the parity bits generated by

G7/G4 are in turn more important than the bits generated by

G5/G4. In order to achieve better performance, these properties are considered in an RM algorithm, i.e. matching the RM algorithm to the channel code used. This can even be done when the generic RM algorithm is employed, e.g. by properly ordering the input bits to the generic RM algorithm and/or properly setting the parameters of the generic RM algorithm.

For example, for 10 code bits generated by G7/G4 and 10 bits by G5/G4 one needs to puncture 7 bits from these 20 parity bits. Since G7/G4 is slightly stronger than G5/G4, it is desirable to puncture 3 bits out of the 10 bits generated by G7/G4, and 4 bits out of the 10 bits generated by G5/G4. This may be done by ordering the input bits to the rate matching algorithm in the way: G7/G4, G5/G4, G1/G4, G5/G4, G7/G4, .... On the other hand, when a generic rate matching algorithm punctures more the G7/G4 bits for the bit sequence G7/G4, G5/G4, G7/G4, G5/G4, G7/G4, .... _r we can reorder the bit sequence as G5/G4, G7/G4, G5/G4, G7/G4, G5/G4, ... such that more G5/G4 bits are punctured than G7/G4 bits.

Generic uniform RM algori thm Notation used (see [5]) : |_xj Round x towards -∞, i.e. integer such that

Absolute value of x.

I Number of TrCHs in the coded composite transport channel (CCTrCH) .

N_data Total number of bits that are available in a radio block for the CCTrCH.

N Number of bits in an encoded block before rate matching on TrCH i with transport format combination j .

ΔN,_. If positive, Δ7Y_1> denotes the number of bits that have to be repeated in a radio segment on TrCH i with transport format combination j in order to produce a radio frame. If negative, ΔN_h] denotes the number of bits that have to be punctured in a radio segment on TrCH i with transport format combination j in order to produce a radio frame .

If null, no bits have to be punctured nor repeated, i.e. the rate matching is transparent and the content of the radio frame is identical to the content of the radio segment on TrCH i with transport format combination j .

RM_τ Semi-static rate matching attribute for transport channel i. e_ιnι Initial value of variable e in the rate matching pattern determination algorithm. GpiusIncrement of variable e in the rate matching pattern determination algorithm.

Sπ_unus Decrement value of variable e in the rate matching pattern determination algorithm. b : Indicates systematic and parity bits

: Systematic bit.

-b=2: 1^st parity bit.

-=3: 2^nd parity bit. Z _J Intermediate calculation variable.

For each radio block using transport format combination j, the number of bits to be repeated or punctured AN_lfJ within one encoded block for each TrCH i is calculated with the following equations: Z_0>, =0

∑RM_m xN_m )xN_data m=1 J

^Z>, = for all i = 1 ... I m=l

Atf _J = Z_t -Z^ -N _J for all i = 1 ... J

If repetition is to be performed, i.e. AN_{l tJ} >0, the following parameters are to be used for each TrCH:

e_pius = 2 N_; »

= 2 x ΔN,

If puncturing is to be performed, the parameters below shall be used. Index b is used to indicate systematic (-=l), 1^st parity ( =2) , and 2^nd parity bit (ib=3) . a=2 when b=2

a=l when b=3

Splui αx

= αx ΔN

The rate matching rule is as follows: if AN, '.7, < 0 — puncturing is to be performed e = e_ιnι — initial error between current and desired puncturing ratio m = 1 — index of current bit do while m < N_{J 3} — for each bit of the radio segment of TrCHi e = e - e_mlnus — update error if e < 0 then — check if bit number should be punctured puncture bit h_xτn — bit is punctured e = e + e_pius — update error end if m = m + 1 — next bit end do else if ΔN, >., > 0 — repetition is to be performed

S ⁼ 6χnj. — initial error between current and desired puncturing ratio m = 1 — index of current bit do while m ≤ N _] for each bit of the radio segment of TrCHi e — e — e_mιnus — update error do while e < 0 check if bit number m should be repeated repeat bit bi,_r — repeat bit s e + pius — update error end do m = m + 1 next bit end do else - ΔN,_,, = 0 do nothing no repetition nor puncturing end if.

Simulation results

Simulations were run in TU3 ideal FH (Typical Urban with Ideal Frequency Hopping at 3 km/h vehicle speed) at 900 MHz carrier frequency for 50 000 speech frames each. A 24 bit coded TFCI was assumed and error detection was provided by a 12-bit CRC. The effective coding rate, i.e. the ratio of the output block size after rate matching to the input block size of the convolutional encoder, was varied from 1 to 0.13.

In terms of Bit Error Rate BER, the RSC code should - according to theory - outperform the ΝRΝSC code for low Carrier-to-interference ratio C/I values but slightly degrade for high C/I values. An additional advantage of an RSC code can be expected for effective coding rates close to 1 (if no systematic bits are punctured) . In the extreme case of effective coding rate = 1, the RSC code offers uncoded transmission (i.e. no coding) which should result in a better performance than NRNSC coding for a certain C/I range. The

C/I intervals where the RSC code achieves a lower BER than the NRNSC code correspond in fact to realistic operating points .

Figures 2 to 6 show the simulation results of BER for four different values of the effective coding rate. For effective coding rates close to 1 (Figures 2 and 3) the RSC code performs much better than the NRNSC code. The performance advantage of the RSC code diminishes at an effective coding rate of 1/3 (Figure 5) , but does not show significant performance degradation for high-repetition scenarios, e.g., an effective coding rate of 1/6 (Figure 6) . The block error rate is identical for all scenarios (not shown here) .

Figure 7 illustrates a system wherein the apparatus and method can be used. A Mobile Station 711 is connected to a mobile network 7° ^v^^a the air interface of the mobile network 70. Typically, for this purpose the mobile station 711 has an antenna 715. Correspondingly, base stations 703 of the mobile network 70 have antennas 705 for communication.

The mobile station 711 has a channel coder 717 and rate matcher 718. Similarly, the base station 703 has channel coder 707 and rate matcher 708. The channel coders 707, 717 and rate matchers 708, 718 ensure that the data bits in the physical layer are interpreted correctly.

Data to be transmitted over the air interface may be transferred via a Radio Network Controller RNC 701 or a Mobile Switching Center MSC 700. The mobile phone 711 can be in communications with other telecommunications terminals, for which the network infrastructure is utilised.

Figure 8 is a block diagram of the channel coder 707, 717 and rate matcher 708, 718, i.e. FLO channel coding and rate matching apparatus 10. In the following the channel coder 707 and rate matcher 708 are discussed, but similar approach applies also for the channel coder 717 and rate matcher 718.

Channel coder 707 performs the channel coding step 102 using convolution means 801 and code generator 803. The convolution means 801 are programmed to perform convolution using a recursive algorithm, preferably a recursive systematic convolutional channel algorithm. The channel coder 707 has also means for passing mode information to rate matcher 708, the mode information including at least some of the properties of the recursive systematic ocnvolutional channel encoder.

The rate matcher 708 includes repeating or puncturing means

811 and adapter 813. The adapter 813 is adapted to adjust the rate matcher algorithm responsive to mode information received from the channel coder 707. The generic rate matching algorithm to be used in the adapter 813 may be Turbo code, or any other code, such as the one to be used for the UTRAN.

Further, the repeating or puncturing means 811 may be instructed not to puncture any or all of the code bits.

Even though only channel encoding has been disclosed, similar approach applies well for channel decoding. Although the invention was described above with reference to the examples shown in the appended drawings, it is obvious that the invention is not limited to these, but it may be modified by those skilled in the art without departing from the scope and the spirit of the invention.

References

[1] TDoc GP-021391 Channel coding for FLO. Nokia. 3GPP TSG GERAN #10.

[2] TDoc GP-020999 Architecture for a Flexible Layer One. Nokia. 3GPP TSG GERAN#9.

[3] TDoc GP-022194 TF for FLO. Ericsson, Nokia, Siemens. 3GPP TSG GERAN #11.

[4] T. Hindelang, J. Hagenauer, M. Schmautz and W. Xu,

"Channel coding techniques for adaptive multirate speech transmission," in Proc. IEEE ICC 00, 2000. [5] 3GPP TS 25.212, ^Multiplexing and channel coding (FDD)".

Claims

1. Method for channel coding (102) and rate matching (103) of information bits (12, 13) comprising:

- encoding (102) first data bits (12) to code bits (13) using a recursive systematic convolutional channel encoding method; and

- rate matching (103) said code bits (13) by repeating or/and puncturing the code bits (13) .

2. Method according to claim 1, wherein: the rate matching step (103) is responsive to the properties of encoding step (102) , in particular to the properties of the recursive systematic convolutional channel encoder.

3. Method according to claim 1 or 2, wherein: the code bits

(13) are not punctured in the rate matching step (103) .

4. Method according to claim 1, 2, or 3, wherein: the rate matching step (103) is performed using an algorithm standardized for UTRAN.

5. Method according to claim 1, 2, 3, or 4, wherein: the rate matching step (103) is performed using a rate matching algorithm for Turbo code.

6. An apparatus (10) for flexible layer one coding and rate matching, characterised in that the apparatus (10) includes : - convolution means (801) , especially corresponding a recursive systematic convolutional channel encoder, adapted to encode information bits (12) ; code generator (803) for generating code bits (13) from said encoded information bits; and

- rate matching means (708) , preferably adapted to utilise a generic rate matching algorithm, adapted to repeat or/and puncture the code bits (13) .

7. Apparatus according to claim 6, wherein: repeating/puncturing means (811) of the rate matching means (708) are functionally connected to an adapter (813) adapted to select an algorithm in response to mode information of the convolution means (801) and/or code generator (803) .

8. Apparatus according to claim 7, wherein: the repeating/puncturing means (811) are adapted not to puncture the code bits (13) .

9. Apparatus according to claim 6, 7, or 8, wherein: the rate matching means correspond to the one standardised for UTRAN.

10. Apparatus according to any one of claim 6 to 9, wherein: the rate matching means (708, 718) are adapted to perform a rate matching algorithm for Turbo code.

11. A mobile terminal (711), especially a GSM/EDGE terminal, wherein: a method according to any one of claims 1-5 or an apparatus according to any one of claims 6-10 is used.

12. A base station (703), especially to be used for GSM/EDGE traffic, wherein: a method according to any one of claims 1-5 or an apparatus according to any one of claims 6-10 is used.