Systems and methods for training sequence selection, transmission and reception

Related Application

0001This application claims the benefit of prior <patcit id="pcit0001" dnum="US61089712A" dnum-type="L"><text>U.S. Provisional Application No. 61/089,712 filed August 18, 2008</text></patcit>.

Field of the Disclosure

0002The disclosure relates to systems and methods for training sequence selection, transmission and reception.

Background

0003Mobile communication systems employ signal processing techniques against the impact of time variant and frequency selective mobile radio channels to improve the link performance. Equalization is used to minimize intersymbol interference (ISI) caused by multipath fading in frequency selective channels. Since the mobile radio channel is random and time varying, an equalizer needs to identify the time-varying characteristics of the mobile channel adaptively through training and tracking. Time division multiplex access (TDMA) wireless systems such as Global System for Mobile communications (GSM) transmit data in fixed-length timeslots, and a training sequence is included in the timeslot (burst), which is designed to allow the receiver to detect timing information and to obtain channel coefficients through channel estimation for further channel equalization.

0004GSM is a successful digital cellular technology being deployed worldwide. Currently, GSM networks provide both voice and data service for billions of subscribers and are still expanding. The access scheme of GSM is TDMA. As illustrated in <figref idref="f0001">Figure 1</figref>, in the 900 MHz frequency band 100, the downlink 102 and uplink 104 are separated, and each has a 25 MHz bandwidth including 124 channels. Carrier separation is 200 kHz. A TDMA frame 106 consists of 8 timeslots 108 corresponding to one carrier frequency. The duration of a timeslot is 577µs. For a normal burst, one GSM timeslot includes 114 data bits, 26 training sequence bits, 6 tail bits, 2 stealing bits, and 8.25 guard period bits. Currently, only one user's speech is transmitted in each timeslot.

0005Eight training sequences for GSM normal bursts are defined in the 3GPP specification (see TS 45.002, "GERAN: Multiplexing and multiple access on the radio path") and are widely used in practice for burst synchronization and channel estimation in current GSM/EDGE Radio Access Network (GERAN) systems.

0006With the increase in the number of subscribers and voice traffic, great pressure is added on GSM operators especially within countries with dense population. In addition, efficient use of hardware and spectrum resource is desired as voice service prices drop. One approach to increasing voice capacity is to multiplex more than one user on a single timeslot.

0007Voice services over Adaptive Multi-user channels on One Slot (VAMOS) (see GP-081949, 3GPP Work Item Description (WID): Voice services over Adaptive Multi-user channels on One Slot) (note: Multi-User Reusing-One-Slot (MUROS) (see GP-072033, "WID": Multi-User Reusing-One-Slot) is the corresponding study item)) is an ongoing work item in GERAN that seeks to increase voice capacity of the GERAN in the order of a factor of two per BTS transceiver both in the uplink and the downlink by multiplexing at least two users simultaneously on the same physical radio resource, i.e., multiple users share the same carrier frequency and the same timeslot. Orthogonal Sub Channel (OSC) (see GP-070214, GP-071792, "Voice capacity evolution with orthogonal sub channel"), co-TCH (see GP-071738, "Speech capacity enhancements using DARP") and Adaptive Symbol Constellation (see GP-080114 "Adaptive Symbol Constellation for MUROS (Downlink)") are three MUROS candidate techniques.

0008In the uplink of OSC, co-TCH, and Adaptive Symbol Constellation two users sharing the same timeslot employ GMSK (Gaussian minimum shift keying) modulation with different training sequences. The base station uses signal processing techniques such as diversity and/or interference cancellation to separate two users' data. Similar to the uplink, in the downlink of co-TCH, two different training sequences are used for DARP (Downlink Advanced Receiver Performance) capable mobiles to separate two users. In the downlink of OSC or Adaptive Symbol Constellation, two subchannels are mapped to the I-and Q-subchannels of a QPSK-type or Adaptive QPSK (AQPSK-type) modulation in which the ratio of I-subchannel and Q-subchannel can be adaptively controlled. Two subchannels use different training sequences as well.

0009<figref idref="f0002">Figure 2</figref> lists eight GSM training sequence codes of 26 bits, each of which has a cyclic sequence structure, i.e., the reference sequence of 16 bits is in the middle and 10 guard bits (5 guard bits are in each side of the reference sequence). The most significant 5 bits and least significant 5 bits of the reference sequence are copied and arranged to append to and precede the reference sequence, respectively. The guard bits can cover the time of intersymbol interference and make the training sequence resistant to time synchronization errors. Each GSM training sequence has ideal periodic autocorrelation properties for non-zero shifts within [-5, 5] when the 16-bit reference sequence is considered only.

0010In GP-070214, GP-071792, "Voice capacity evolution with orthogonal sub channel", a new set of eight training sequences of length 26 bits was proposed for OSC, in which each of new training sequences is optimized in cross-correlation properties with the corresponding legacy GSM training sequence. The new sequences are listed in <figref idref="f0002">Figure 3</figref>. It can be observed that these new training sequences do not preserve the cyclic sequence structure as the legacy GSM training sequences.

0011In GP-080785 and GP-080641, "New series of training sequence codes for MUROS", new training sequence codes are proposed. The new TSCs of GP-080785 have a central 16 bits containing the total information of the 26 bit TSC. The remaining 10 bits are set in head and tail position for protection. Further, it is described that "the design principles are as follows: Firstly, the new TSC must have low cross-correlation with the existing one that paired with it. Secondly, the new TSC is desired to have low cross-correlation with other new TSCs. Thirdly, it's required that new TSCs have good correlation property in general with all existing training sequences, not only with its pair" <i>(sic).</i> Additionally, GP-070620, "New training sequences for RED HOT and HUGE", describes that "for RED HOT and HUGE work items, new training sequence codes (TSCs) should be considered to have both good autocorrelation and cross-correlation properties to improve channel estimation". GP-070707, "Training Sequences for High Symbol Rate", similarly discloses training sequences for RED HOT and HUGE which are evaluated and compared with other proposed training sequences.

Summary

0012Various aspects of the present invention are set out in the appended claims.

Brief Description of the Drawings

0013Embodiments of the application will now be described with reference to the attached drawings in which: <ul id="ul0001" list-style="none" compact="compact"><li><figref idref="f0001">Figure 1</figref> is a schematic diagram of a bandwidth allocation and TDMA frame definitions for GSM;</li><li><figref idref="f0002">Figure 2</figref> is a table listing the legacy GSM training sequences;</li><li><figref idref="f0002">Figure 3</figref> is a table containing a set of training sequences with optimized cross-correlation properties compared to the legacy GSM training sequences;</li><li><figref idref="f0003">Figure 4A</figref> is a table containing a set of training sequences;</li><li><figref idref="f0003">Figure 4B</figref> is a schematic diagram of a computer readable medium containing the training sequences of <figref idref="f0003">Figure 4A</figref>;</li><li><figref idref="f0004">Figure 5A</figref> is a table containing a set of training sequences;</li><li><figref idref="f0004">Figure 5B</figref> is a schematic diagram of a computer readable medium containing the training sequences of <figref idref="f0004">Figure 5A</figref>;</li><li><figref idref="f0005">Figure 6A</figref> is a table containing a set of training sequences;</li><li><figref idref="f0005">Figure 6B</figref> is a schematic diagram of a computer readable medium containing the training sequences of <figref idref="f0005">Figure 6A</figref>;</li><li><figref idref="f0006">Figure 7</figref> depicts several sets used to define a set of training sequences;</li><li><figref idref="f0007">Figure 8</figref> is a flowchart of a first method of determining training sequences;</li><li><figref idref="f0008">Figure 9A</figref> is a flowchart of a first method of assigning training sequences;</li><li><figref idref="f0008">Figure 9B</figref> is a flowchart of a second method of assigning training sequences;</li><li><figref idref="f0009">Figure 10A</figref> is a block diagram of a transmitter for OSC downlink transmission;</li><li><figref idref="f0010">Figure 10B</figref> is a block diagram showing a pair of receivers of OSC subchannels;</li><li><figref idref="f0011">Figure 11A</figref> is a block diagram of transmitter of co-TCH for downlink transmission;</li><li><figref idref="f0012">Figure 11B</figref> is a block diagram of a pair of receivers of co-TCH downlink transmission;</li><li><figref idref="f0013">Figure 12A</figref> shows a pair of transmitters of OSC or co-TCH for uplink transmission; and</li><li><figref idref="f0014">Figure 12B</figref> is a block diagram of a receiving apparatus composed of two receivers for receiving respective transmissions from the pair of transmitters of <figref idref="f0013">Figure 12A</figref>.</li></ul>

Detailed Description

0014The degradation of signal-to-noise ratio (SNR) (see <nplcit id="ncit0001" npl-type="s"><text>B. Steiner and P. Jung, "Optimum and suboptimum channel estimation for the uplink CDMA mobile radio systems with joint detection", European Transactions on Telecommunications, vol. 5, Jan.-Feb., 1994, pp. 39-50</text></nplcit>, and M. Pukkila and P. Ranta, "Channel estimator for multiple co-channel demodulation in TADM mobile systems", Proc. of the 2nd EPMC, Germany) is used herein to evaluate the correlation properties of training sequences and/or to design new training sequences. In MUROS/VAMOS, the interference comes from the other subchannel of the same MUROS/VAMOS pair in the same cell and also from co-channel signals of other cells.

0015The degradation in SNR can be determined as follows. Let a training sequence of length N be S = {<i>s</i>₁,<i>s</i>_2,...,<i>s_N</i>}<i>, s_n</i> ∈ {-1,+1}, <i>n</i> =1,..., <i>N.</i> Consider two synchronous co-channel or MUROS/VAMOS signals with L-tap independent complex channel impulse responses <i>h_m</i> = <i>(h</i>_<i>m,</i>1, <i>h</i>_<i>m</i>,2<i>,..., h_m,_L), m</i> =1, 2. The joint channel impulse response is <i>h</i> = (<i>h</i>₁,<i>h</i>₂). Let the received signal samples at the receiver be: <i>y</i> = <i>Sh' + n</i> where the noise vector is <i>n</i> = (<i>n₁,n₂,...n</i>_<i>N-L</i>+1)' and S = [S₁,S₂] is a <i>(N - L+ 1)</i> × 2<i>L</i> matrix and S_m ( <i>m</i> =1, 2) is defined as below <maths id="math0001" num="(1)"><math display="block"><msub><mi>S</mi><mi>m</mi></msub><mo>=</mo><mfenced open="[" close="]"><mtable><mtr><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mi>L</mi></mrow></msub></mtd><mtd><mo>⋯</mo></mtd><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mn>2</mn></mrow></msub></mtd><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mn>1</mn></mrow></msub></mtd></mtr><mtr><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mi>L</mi><mo>+</mo><mn>1</mn></mrow></msub></mtd><mtd><mo>⋯</mo></mtd><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mn>3</mn></mrow></msub></mtd><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mn>2</mn></mrow></msub></mtd></mtr><mtr><mtd><mo>⋮</mo></mtd><mtd><mo>⋮</mo></mtd><mtd><mo>⋮</mo></mtd><mtd><mo>⋮</mo></mtd></mtr><mtr><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mi>N</mi></mrow></msub></mtd><mtd><mo>⋯</mo></mtd><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mi>N</mi><mo>-</mo><mi>L</mi><mo>+</mo><mn>2</mn></mrow></msub></mtd><mtd><msub><mi>s</mi><mrow><mi>m</mi><mo>,</mo><mi>N</mi><mo>-</mo><mi>L</mi><mo>+</mo><mn>1</mn></mrow></msub></mtd></mtr></mtable></mfenced></math><img file="EP2157752B1_D0001.tif" /></maths> which is correspondent to the training sequence (<i>s_m,1, s_m.2,...,s_m,N</i>) (note that S₁ and S₂ can be constructed with two different training sequences, respectively, either from the same training sequence set or from different training sequence sets).

0016The least-squared error estimate of the channel is: <maths id="math0002" num="(2)"><math display="block"><msup><mover><mi>h</mi><mo>^</mo></mover><mi>t</mi></msup><mo>=</mo><msup><mfenced><msup><mi>S</mi><mi>t</mi></msup><mo>⁢</mo><mi>S</mi></mfenced><mrow><mo>-</mo><mn>1</mn></mrow></msup><mo>⁢</mo><msup><mi>S</mi><mi>t</mi></msup><mo>⁢</mo><mi>y</mi><mn>.</mn></math><img file="EP2157752B1_D0002.tif" /></maths> The SNR degradation of training sequences is defined as: <maths id="math0003" num="(3)"><math display="block"><msub><mi>d</mi><mi mathvariant="italic">SNR</mi></msub><mo>=</mo><mn>10</mn><mo>⋅</mo><msub><mi>log</mi><mn>10</mn></msub><mo>⁢</mo><mfenced><mn>1</mn><mo>+</mo><mi mathvariant="italic">tr</mi><mo>⁢</mo><mfenced open="[" close="]"><msup><mfenced><msup><mi>S</mi><mi>t</mi></msup><mo>⁢</mo><mi>S</mi></mfenced><mrow><mo>-</mo><mn>1</mn></mrow></msup></mfenced></mfenced><mspace width="1em" /><mfenced><mi>dB</mi></mfenced></math><img file="EP2157752B1_D0003.tif" /></maths> where <i>tr</i>[<i>X</i>] is the trace of matrix <i>X</i> and <i>Q</i> = [<i>q_ij</i>]_{2<i>L</i>×2<i>L</i>} = <i>S'S</i> is a correlation matrix including the autocorrelations of S₁ and S₂, and cross-correlation between S₁ and S₂ with calculation of entries as: <maths id="math0004" num="(4)"><math display="block"><msub><mi>q</mi><mi mathvariant="italic">ij</mi></msub><mo>=</mo><mrow><mo>{</mo><mtable columnalign="left"><mtr><mtd><mstyle displaystyle="true"><munderover><mo>∑</mo><mrow><mi>n</mi><mo>=</mo><mn>1</mn></mrow><mrow><mi>N</mi><mo>-</mo><mi>L</mi><mo>+</mo><mn>1</mn></mrow></munderover></mstyle><msub><mi>s</mi><mrow><mn>1</mn><mo>,</mo><mi>n</mi><mo>+</mo><mi>L</mi><mo>-</mo><mi>i</mi></mrow></msub><mo>⁢</mo><msub><mi>s</mi><mrow><mn>1</mn><mo>,</mo><mi>n</mi><mo>+</mo><mi>L</mi><mo>-</mo><mi>j</mi></mrow></msub><mo>,</mo></mtd><mtd><mi mathvariant="italic">if i</mi><mo>≤</mo><mi>L</mi><mo>,</mo><mi>j</mi><mo>≤</mo><mi>L</mi></mtd></mtr><mtr><mtd><mstyle displaystyle="true"><munderover><mo>∑</mo><mrow><mi>n</mi><mo>=</mo><mn>1</mn></mrow><mrow><mi>N</mi><mo>-</mo><mi>L</mi><mo>+</mo><mn>1</mn></mrow></munderover></mstyle><msub><mi>s</mi><mrow><mn>1</mn><mo>,</mo><mi>n</mi><mo>+</mo><mi>L</mi><mo>-</mo><mi>i</mi></mrow></msub><mo>⁢</mo><msub><mi>s</mi><mrow><mn>2</mn><mo>,</mo><mi>n</mi><mo>+</mo><mn>2</mn><mo>⁢</mo><mi>L</mi><mo>-</mo><mi>j</mi></mrow></msub><mo>,</mo></mtd><mtd><mi mathvariant="italic">if i</mi><mo>≤</mo><mi>L</mi><mo>,</mo><mi>L</mi><mo>&lt;</mo><mi>j</mi><mo>≤</mo><mn>2</mn><mo>⁢</mo><mi mathvariant="italic">L or j</mi><mo>≤</mo><mi>L</mi><mo>,</mo><mi>L</mi><mo>&lt;</mo><mi>i</mi><mo>≤</mo><mn>2</mn><mo>⁢</mo><mi>L</mi><mn>.</mn></mtd></mtr><mtr><mtd><mstyle displaystyle="true"><munderover><mo>∑</mo><mrow><mi>n</mi><mo>=</mo><mn>1</mn></mrow><mrow><mi>N</mi><mo>-</mo><mi>L</mi><mo>+</mo><mn>1</mn></mrow></munderover></mstyle><msub><mi>s</mi><mrow><mn>2</mn><mo>,</mo><mi>n</mi><mo>+</mo><mn>2</mn><mo>⁢</mo><mi>L</mi><mo>-</mo><mi>i</mi></mrow></msub><mo>⁢</mo><msub><mi>s</mi><mrow><mn>2</mn><mo>,</mo><mi>n</mi><mo>+</mo><mn>2</mn><mo>⁢</mo><mi>L</mi><mo>-</mo><mi>j</mi></mrow></msub><mo>,</mo></mtd><mtd><mi mathvariant="italic">if L</mi><mo>&lt;</mo><mi>i</mi><mo>≤</mo><mn>2</mn><mo>⁢</mo><mi>L</mi><mo>,</mo><mi>L</mi><mo>&lt;</mo><mi>j</mi><mo>≤</mo><mn>2</mn><mo>⁢</mo><mi>L</mi></mtd></mtr></mtable></mrow></math><img file="EP2157752B1_D0004.tif" /></maths>

0017Based on definitions (1)-(3), the pairwise SNR degradation values between GSM training sequences are calculated and listed in Table 1. <tables id="tabl0001" num="0001"><img file="EP2157752B1_D0005.tif" /></tables>

0018The average, minimum and maximum pairwise SNR degradation values between different GSM training sequences equal 5.10 dB, 2.72 dB and 11.46 dB, respectively. Table 1 demonstrates that some GSM training sequence pairs result in reasonable SNR degradation values while some GSM training sequence pairs are strongly correlated. It seems not to be suitable to apply all existing GSM training sequences to MUROS/VAMOS. It would be desirable to have new training sequences for MUROS/VAMOS, each having very good autocorrelation properties and very good cross-correlation properties with the corresponding GSM training sequence. It would also be desirable to reduce the effects of co-channel interference, cross-correlation properties for any pairs of new training sequences and cross-correlation properties for any pairs of new training sequences and legacy GSM training sequences through further optimization.

0019Tables 2 and 3 present the pairwise SNR degradation performance of the sequences of <figref idref="f0002">Figure 3</figref> between any pairs of these sequences and GSM training sequences, and between any pairs of these sequences themselves. <tables id="tabl0002" num="0002"><img file="EP2157752B1_D0006.tif" /></tables><tables id="tabl0003" num="0003"><img file="EP2157752B1_D0007.tif" /></tables><tables id="tabl0004" num="0004"><img file="EP2157752B1_D0008.tif" /></tables>

0020In Table 2, the pairwise SNR degradation values in the diagonal of the table are the results of a sequence of <figref idref="f0002">Figure 3</figref> and a corresponding GSM training sequences. In this document, the corresponding sequences are defined as two sequences with the same training sequence number in two separate sequence tables. The average of the diagonal values in Table 2 equals 2.11 dB. The average, minimum and maximum SNR degradation values between any pairs of sequences of <figref idref="f0002">Figure 3</figref> and GSM TSCs are 2.63 dB, 2.05 dB and 4.87 dB, respectively.

0021Table 3 shows that the average, minimum and maximum SNR degradation values between any pairs of different sequences of <figref idref="f0002">Figure 3</figref> are 3.19 dB, 2.32 dB and 6.89 dB, respectively.

0022Both Tables 2 and 3 demonstrate that the average pairwise SNR degradation performance between any pairs of sequences of Table 2 and GSM training sequences, and any pairs of different sequences of Table 2 is good. However, the peak pairwise SNR degradation values shown in Table 2 and 3 may affect co-channel interference cancellation with the introduction of MUROS/VAMOS.

New Training Sequences for MUROS/VAMOS

A. Training sequences best-paired with the corresponding GSM TSCs

0023In an embodiment of the disclosure, a set of eight sequences of length 26 are obtained, through computer search, which are best-paired with the corresponding GSM training sequences, respectively, in terms of SNR degradation calculated with (1)-(3). <figref idref="f0003">Figure 4A</figref> shows these best-paired sequences, referred to as Training Sequence Set A, generally indicated at 120 in a data structure 122 stored on a computer readable medium 124. The search was conducted as follows: <ol id="ol0001"><li>1) start with first GSM training sequence;</li><li>2) exhaustively search through set of all candidate sequences for the sequence with the lowest SNR degradation, and add the sequence found to the new set, and remove the sequence found from the candidate set;</li><li>3) repeat steps 1 and 2 for sequences that are best paired with each of the second through eighth GSM training sequences.</li></ol>

0024Shown in <figref idref="f0003">Figure 4B</figref> is a computer readable medium generally indicated at 128 upon which is stored a data structure 125. The data structure 125 includes the set of standard GSM training sequences 126, and includes the training sequence set A 127. There is a one to one correspondence between the GSM training sequences 126 and the training sequence set A 127. <tables id="tabl0005" num="0005"><img file="EP2157752B1_D0009.tif" /></tables>

0025The average, minimum and maximum SNR degradation values between any pairs of sequences in <figref idref="f0003">Figure 4A</figref> and GSM training sequences are 2.52 dB, 2.04 dB and 4.10 dB, respectively. The average of the diagonal values in Table 4 equals 2.07 dB. Based on results shown in Table 4, the new training sequences of <figref idref="f0003">Figure 4A</figref> are well-designed to be paired with the corresponding GSM training sequences.

0026Table 5 demonstrates SNR degradation values between sequences listed in <figref idref="f0003">Figure 4A</figref>. The average, minimum and maximum pairwise SNR degradation values between sequences best-paired with GSM training sequences are 3.04 dB, 2.52 dB and 4.11 dB, respectively. <tables id="tabl0006" num="0006"><img file="EP2157752B1_D0010.tif" /></tables>

B. Training sequences with cyclic structure with optimized autocorrelation and cross-correlation properties

0027A set of training sequences with optimized autocorrelation and cross-correlation properties was determined by computer search using a method described in detail below. The set of training sequences is set out in <figref idref="f0004">Figure 5A</figref>, referred to as Training Sequence Set B, generally indicated at 130, in a data structure 132 stored on a computer readable medium 134.

0028Shown in <figref idref="f0004">Figure 5B</figref> is a computer readable medium generally indicated at 138 upon which is stored a data structure 135. The data structure 135 includes the set of standard GSM training sequences 136, and includes the training sequence set B 137. There is a one to one correspondence between the GSM training sequences 136 and the training sequence set A 137.

0029The pairwise SNR degradation values between any pairs of new training sequences in <figref idref="f0004">Figure 5A</figref> and GSM training sequences are shown in Table 6. The average, minimum and maximum SNR degradation values in Table 6 are 2.43 dB, 2.13 dB and 2.93 dB, respectively. The average of the diagonal values in Table 6 equals 2.22 dB. <tables id="tabl0007" num="0007"><img file="EP2157752B1_D0011.tif" /></tables>

0030Table 7 demonstrates pairwise SNR degradation values between sequences listed in <figref idref="f0004">Figure 5A</figref>. The average, minimum and maximum pairwise SNR degradation values in Table 7 are 3.17 dB, 2.21 dB and 4.75 dB, respectively. <tables id="tabl0008" num="0008"><img file="EP2157752B1_D0012.tif" /></tables>

C. Training Sequences without cyclic structure

0031Unlike training sequence set B, a third training sequence set, referred to herein as training sequence set C is composed of sequences that do not maintain cyclic structure. Only optimization procedure II-IV for training sequence set B outlined below are taken into account for generation of training sequence set C. For <i>optimization</i> of <i>SNR degradation between new</i> sequences <i>and GSM training sequences,</i> the sequence set Ω1 is obtained by selecting |Ω1| sequences from 2²⁶ sequences with minimum average SNR degradation values between sequences in |Ω1| and all GSM training sequences. Training sequence set C is listed in <figref idref="f0005">Figure 6A</figref>, generally indicated at 140 in a data structure 142 stored on a computer readable medium 144.

0032Shown in <figref idref="f0005">Figure 6B</figref> is a computer readable medium generally indicated at 148 upon which is stored a data structure 145. The data structure 145 includes the set of standard GSM training sequences 146, and includes the training sequence set C 147. There is a one to one correspondence between the GSM training sequences 146 and the training sequence set C 147.

0033The pairwise SNR degradation values between any pairs of new training sequences in <figref idref="f0005">Figure 6A</figref> and GSM training sequences are shown in Table 8. The average, minimum and maximum SNR degradation values in Table 8 are 2.34 dB, 2.11 dB and 2.87 dB, respectively. The average of the diagonal values in Table 8 equals 2.16 dB. <tables id="tabl0009" num="0009"><img file="EP2157752B1_D0013.tif" /></tables>

0034Table 9 demonstrates pairwise SNR degradation values between sequences listed in <figref idref="f0005">Figure 6A</figref>. The average, minimum and maximum pairwise SNR degradation values in Table 9 are 3.18 dB, 2.44 dB and 4.19 dB, respectively. <tables id="tabl0010" num="0010"><img file="EP2157752B1_D0014.tif" /></tables>

Sequence search procedure - First Method

0035<figref idref="f0006">Figure 7</figref> illustrates a procedure for finding a second set of |Ω3| training sequences Ω3 that has auto-correlation and cross-correlation properties having regard to a target set of training sequences Ψ. The procedure involves determining sequence sets Ω 152, S21 154,02 156 and Ω3 158 where Ω ⊃ Ω1 ⊃ Ω2 ⊃ Ω3 and | Ω3 | = the number of sequences to be found, where | · | represents the number of elements in a set. Ω 152 is a subset of the set of all possible sequences 150 determined through a first optimization step. Ω1 154 is a subset of Ω2 152 determined through a second optimization step. Ω2 156 is a subset of Ω1 154 determined through a third optimization step. Ω3 158 is a subset of Ω2 156 determined through a fourth optimization step.

0036The method is computer implemented and will be described with reference to the flowchart of <figref idref="f0007">Figure 8</figref>. The method begins at block 8-1 with the optimization of autocorrelations for a candidate set of training sequences to produce a first training sequence set. The method continues at block 8-2 with optimization of SNR degradation between sequences of the first training sequence set and a target set of training sequences Ψ to produce a second training sequence set. The method continues at block 8-3 with optimization of SNR degradation among sequences of the second training sequence set to produce a third training sequence set. The method continues at block 8-4 with the optimization of SNR degradation between training sequences of the third set and the corresponding sequences of the target set of training sequences Ψ. The output of block 8-4 is a fourth set of training sequences, and this is the new training sequence set. This set is output for use in a system with multi-user transmission. Another embodiment provides a computer readable medium having instructions stored thereon which, when executed by a computer, cause the method of <figref idref="f0007">Figure 8</figref> to be performed.

0037In some embodiments, steps 8-2 and 8-3 are performed in the reverse order to that shown and described above. This results in a computer method comprising: optimizing cross-correlations between sequences among a first training sequence set to produce a second training sequence set; optimizing cross-correlations between sequences of the second training sequence set and a target training sequence set to produce a third training sequence set; optimizing cross-correlations between sequences of the third training sequence set and corresponding sequences of the target training sequence set to produce a fourth training sequence set; and outputting the fourth training sequence set for use in a multi-user transmission system.

0038First Optimization step: <i>optimization</i> of <i>autocorrelations:</i> Consider all binary sequences of a desired length. Optionally copy some of the last bits of the sequence onto the front to make the sequence somewhat cyclic in nature. Search for sequences with zero autocorrelation values for a range of non-zero shifts.

0039In order to achieve zero autocorrelations, the sequences upon which autocorrelation are determined must have even length. If an odd length sequence is required, an additional bit is added to the set of sequences with zero autocorrelation values. This might for example be done by copying the first bit of the original sequence, or by appending -1 or +1 for further optimization of cross-correlation properties.

0040The output of this step is a set of sequences, Ω, with optimized autocorrelation properties. For sequences to which an additional bit was added, the maximum magnitude of autocorrelation coefficients would be 1 in correlation matrix <i>Q = S'S</i> in (3);

0041Second Optimization Step: <i>optimization of SNR degradation between new sequences and target</i> set of <i>training sequences</i> Ψ : a subset of Ω, Ω1, is obtained by selecting |Ω1| sequences from Ω with minimum average SNR degradation values between sequences in |Ω1| and the target set of training sequences Ψ. The average SNR degradation for a given sequence from Ω is determined by computing the degradation for that sequence and each of the target set training sequences and averaging the result.

0042Third Optimization Step: <i>optimization</i> of <i>SNR degradation between new sequences:</i> a subset of Ω1, Ω2, with minimum average SNR degradation values between sequences in |Ω2| is selected. The following is an example of how the third optimization step might be performed: <ol id="ol0002"><li>1) pick a first sequence from Ω2 and remove from Ω2;</li><li>2) examine all remaining sequences in Ω2 for the one with the lowest SNR degradation with the first sequence and select that as the second sequence, and remove from Ω2;</li><li>3) examine all remaining sequences in Ω2 for the one with the lowest average SNR degradation with the first sequence and the second sequence, and remove from Ω2;</li><li>4) and so on until a desired number of sequences have been identified. Calculate the average SNR degradation between the sequences thus identified;</li><li>5) repeat steps 1 to 4 using a different first sequence from Ω2 to generate a respective set of sequences and a respective average SNR degradation;</li><li>6) of all the sets of sequences thus generated, pick the set of sequences with the minimum average SNR degradation.</li></ol>

0043Fourth Optimization Step: <i>optimization of SNR degradation between new training sequences and the corresponding sequences of the target</i> set <i>of training sequences</i> Ψ: |Ω3| sequences out of the sequence set Ω2 are selected. This step is used to determine pairs of training sequences that include one from the target set and one from the new set. The following is an example approach to performing this step: <ol id="ol0003"><li>a) select a training sequence from the target set;</li><li>b) find the training sequence in the new set that has the lowest SNR degradation with the training sequence of the target set, and pair that training sequence with the first training sequence from the target set, and remove that training sequence from the set of available training sequences;</li><li>c) repeat steps a) and b) until all sequences from the target set have been selected.</li></ol>

0044The above-described optimization procedure was applied to develop the set of training sequences in <figref idref="f0004">Figure 5A</figref>, with |Ω1|=120 and |Ω2|=12. More specifically: <ul id="ul0002" list-style="none" compact="compact"><li>First Optimization Step: <i>optimization of autocorrelations:</i> Consider all binary sequences of length 20 (set size is 2²⁰). Similar to GSM TSCs, for each of such sequences, copy the last of 5 bits of the sequence and precede these 5 bits at the most significant positions to generate a sequence of length 25; search sequences of length 25 with zero autocorrelation values for non-zero shift [-5, 5] by using the autocorrelation definition <maths id="math0005"><math display="inline"><mi>R</mi><mfenced><mi>k</mi></mfenced><mo>=</mo><mstyle displaystyle="true"><munderover><mo>∑</mo><mrow><mi>n</mi><mo>=</mo><mn>6</mn></mrow><mn>25</mn></munderover></mstyle><msub><mi>s</mi><mi>n</mi></msub><mo>⁢</mo><msub><mi>s</mi><mrow><mi>n</mi><mo>+</mo><mi>k</mi></mrow></msub><mo>,</mo><mi>k</mi><mo>=</mo><mo>-</mo><mn>5</mn><mo>,</mo><mo>…</mo><mo>,</mo><mo>-</mo><mn>1</mn></math><img file="EP2157752B1_D0015.tif" /></maths>. There are totally 5440 such sequences available.</li></ul>

0045To be compatible with the current TSC format, the new TSCs length must be 26. The 26th bit of the full-length (length 26) sequences could be obtained either by copying the first bit of the corresponding sequences of length 20 or by appending -1 or +1 for further optimization of cross-correlation properties. Therefore, the set of sequences, Ω, with optimized autocorrelation properties is generated. Both methods will limit the maximum magnitude of autocorrelation coefficients to be 1 in correlation matrix <i>Q = S'S</i> in (3).

0046Second Optimization Step: <i>optimization of SNR degradation between new sequences and GSM TSCs</i>: a subset of Ω, Ω1, are obtained by selecting |Ω1| sequences from Ω with minimum average SNR degradation values between sequences in |Ω1| and all GSM TSCs. The average SNR degradation for a given sequence from Ω is determined by computing the degradation for that sequence and each of the GSM sequences and averaging the result.

0047Third Optimization Step: <i>optimization of SNR degradation between new sequences:</i> a subset or Ω1,Ω2, with minimum average SNR degradation values between sequences in |Ω2| are selected.

0048Fourth Optimization Step: <i>optimization</i> of <i>SNR degradation between new training sequences and the corresponding GSM TSCs</i>: |Ω3|= 8 sequences out of the sequence set | Ω3 | are determined. The result is the set B of sequences <figref idref="f0004">Figure 5A</figref>.

Sequence search procedure - Second Method

0049In another method of sequence search, a search approach that is similar to the above-described 'first method' is provided in which the first optimization step is omitted. In this case, the method begins with the second optimization step of the first method, and the sequence sex Ω1 is obtained by selecting | Ω1 | sequences from all possible sequences with minimum average SNR degradation values between sequences in | Ω1 | and all the sequences of the target set Ψ. Note that the sequences are not required to be cyclic in this embodiment. Using the language of the flowchart of <figref idref="f0007">Figure 8</figref>, block 8-1 is omitted, and the "first training sequence set" becomes the candidate set of training sequences.

0050Applied to the MUROS/VAMOS problem, in the second optimization step, the sequence set Ω1 is obtained by selecting |Ω1| sequences from all 2²⁶ possible sequences of length 26 with minimum average SNR degradation values between sequences in | Ω1 | and all the GSM training sequences. The result is the set of sequences C of <figref idref="f0005">Figure 6A</figref> above.

ASSIGNMENT OF TRAINING SEQUENCES

0051Having defined a new set of training sequences or a part of a new set of training sequences for use in conjunction with a target set of training sequences, for example the set A defined above or a part of the set A in conjunction with legacy GSM training sequences, set B defined above or a part of the set B in conjunction with legacy GSM training sequences or set C defined above or a part of set C in conjunction with legacy GSM training sequences, various mechanisms are provided for assigning training sequences. Note these mechanisms are not specific to the examples provided herein. A specific example of multi-user operation is the above described MUROS/VAMOS operation, for example, specific implementations of which include the OSC or co-TCH or Adaptive Symbol Constellation implementations thereof.

0052In a cell within which multi-user transmission is being implemented, in interference limited scenarios there is interference from at least two sources. This includes interference from the other user(s) on the same physical transmission resource within the cell, and interference from mobile stations of the same physical transmission resource in other cells. Conventional mobile stations are already equipped to deal with the interference from mobile stations using the same physical transmission resource in other cells.

0053A mobile station that is specifically aware of multi-user operation will be referred to as "multi-useraware". In a specific example, a mobile station that is aware of VAMOS aware operation might, for example, be referred to as a VAMOS aware mobile station. Such mobile devices are configured to be able to use any training sequence of the target set and any training sequence of the new set. Mobile stations that are not specifically aware of multi-user operation will be referred to as "multi-user unaware". Such mobile devices are configured to be able to use only training sequences of the target set. Note that multi-user unaware mobile stations may still be served in a multi-user context; such a mobile station will treat the interference from other user(s) on the same physical transmission resource in the same cell in the same manner as it treats mobile stations using the same physical transmission resource in other cells.

0054Similarly, networks may or may not have multi-user capability. A network that has multi-user capability functions using the target set and the new set of training sequences, while a network that does not have multi-user capability uses only the target set of training sequences.

0055In some embodiments, the assignment of training sequences to base stations is done during network configuration, and does not change until a reconfiguration is performed. A given multi-user aware network element such as a base station is configured with a training sequence from the target set and a training sequence from the new set. In the event a base station performs multi-carrier transmission, the base station is configured with a respective training sequence from the target set and a respective training sequence from the new set for each carrier frequency that it uses. The training sequence from the new set is the training sequence that is best paired with the training sequence of the target set and vice versa. In such a case, the training sequences assigned to the mobile stations will be a function of the previously performed network configuration. As a mobile station moves between coverage areas, the training sequences assigned will change. In some embodiments, the same training sequence is assigned for a given mobile station for both uplink transmission and downlink transmission. In other embodiments, different training sequences may be assigned.

0056The behaviour of the networks can be divided into two types: behaviour when a timeslot is to be used for only a single user, and behaviour when a timeslot is to be used for multiple users.

0057Behaviour when a timeslot is to be used for a single user: <ol id="ol0004"><li>A) When a multi-user aware MS is to be served by a network without multi-user capability, one training sequence from the target set will be assigned for this MS, namely the training sequence allocated to the serving base station during network configuration or otherwise. A multi-user aware mobile station is made aware of and capable of using both the target set of training sequences and the new set of training sequences.</li><li>B) When a multi-user aware MS is to be served by a network with multi-user capability, if there is a vacant timeslot, this MS does not need to share a timeslot with another MS. Since on average new training sequences have been designed to have better correlation properties than training sequences of the target set, one new training sequence will be assigned for this MS, namely the new training sequence allocated to the serving base station during network configuration or otherwise.</li></ol>

0058Behaviour when a timeslot is to be used for multiple users <ul id="ul0003" list-style="none" compact="compact"><li>A) When a first multi-user aware MS, MS-A, is served by a network with multi-usercapability, as discussed above, a new training sequence is assigned to MS-A, namely the new training sequence allocated to the serving base station during network configuration or otherwise. If there is a request to share the same timeslot with a second MS, MS-B, no matter whether MS-B is a multi-user aware MS or not, the training sequence of the target set which is best-paired with the new training sequence being used by MS-A will be assigned to MS-B, namely the training sequence of the target set allocated to the serving base station during network configuration or otherwise.</li><li>B) When a first multi-user unaware MS, MS-A, is served by a network with multi-user capability, a training sequence from the target set is assigned to MS-A, namely the training sequence from the target set allocated to the serving base station during network configuration or otherwise. If there is a request to share MS-A with a second MS that is multi-user aware, MS-B, in the same timeslot, the new training sequence which is best-paired with the training sequence of the target set being used by MS-A will be assigned to MS-B, namely the new training sequence allocated to the serving base station during network configuration or otherwise.</li></ul>

0059Two flowcharts of an example method of training sequence assignment using multi-user slots are shown in <figref idref="f0008">Figures 9A and 9B. Figure 9A</figref> relates to case A) described above, while <figref idref="f0008">Figure 9B</figref> relates to case B) described above.

0060Referring now to <figref idref="f0008">Figure 9A</figref>, when a first multi-user aware mobile station is to share with a second mobile station, block 9A-1 involves assigning a new training sequence to the first mobile station. Block 9A-2 involves assigning to the second mobile station the training sequence of the target set which is best-paired with the new training sequence being used by first mobile station, irrespective of whether the second mobile station is a multi-user aware mobile station or not;

0061Referring now to <figref idref="f0008">Figure 9B</figref>, when a first multi-user unaware mobile station is to share with a second mobile station that is multi-user aware, block 9B-1 involves assigning a training sequence from the target set to the first mobile station. Block 9B-2 involves assigning to the second mobile station the training sequence of the new set which is best-paired with the training sequence being used by the first mobile station.

Example Transmitter and Receiver Implementations

0062Various detailed example transmitter and receiver implementations will now be described. <figref idref="f0009">Figure 10A</figref> shows a transmitter of OSC (orthogonal subchannels) (or Adaptive Symbol Constellation) for downlink transmission. Most of the components are standard and will not be described in detail. The transmitter includes a training sequence repository 200 containing a target training sequence set and a new training sequence set generated using one of the methods described above. Typically, during network configuration or otherwise, a training sequence from the target set and a training sequence from the new set are assigned to the transmitter. In some embodiments, for multi-user slots, the training sequences thus assigned are used in accordance with the method of <figref idref="f0008">Figure 9A or 9B</figref> to name a few specific examples. The remainder of the drawing is a specific example of a signal generator for generating a multi-user signal.

0063<figref idref="f0010">Figure 10B</figref> shows a pair of receivers of OSC (orthogonal subchannels) for downlink transmission. Most of the components are standard and will not be described in detail. Each receiver includes a respective memory 210 containing a target training sequence set and a new training sequence set generated using one of the methods described above. One of these, referred to as training sequence A, is used by the top receiver to perform timing/channel estimation and DARP processing, and another of these, referred to as training sequence B, is used by the bottom receiver to perform timing/channel estimation and DARP processing. The training sequences used by the receivers are assigned by the network, and must match up with the training sequences transmitted. When a mobile station moves to a different coverage area to which different training sequences have been assigned, the mobile station changes training sequence it uses accordingly.

0064<figref idref="f0011">Figure 11A</figref> shows a transmitter of co-TCH (co-traffic channel) for downlink transmission. Most of the components are standard and will not be described in detail. The transmitter includes a training sequence repository 200 containing a target training sequence set and a new training sequence set generated using one of the methods described above. Typically, during network configuration or otherwise, a training sequence from the target set and a training sequence from the new set are assigned to the transmitter. In some embodiments, for multi-user slots, the training sequences thus assigned are used in accordance with the method of <figref idref="f0008">Figure 9A or 9B</figref> to name a few specific examples. The remainder of the drawing is a specific example of a signal generator for generating a multi-user signal.

0065<figref idref="f0012">Figure 11B</figref> shows a pair of receivers of co-TCH (co-traffic channel) for downlink transmission. Most of the components are standard and will not be described in detail. Each receiver includes a respective memory 210 containing a target training sequence set and a new training sequence set generated using one of the methods described above. One of these, referred to as training sequence A, is used by the top receiver to perform timing/channel estimation and DARP processing, and another of these, referred to as training sequence B, is used by the bottom receiver to perform timing/channel estimation and DARP processing. The training sequences used by the receivers are assigned by the network, and must match up with the training sequences transmitted. When a mobile station moves to a different coverage area to which different training sequences have been assigned, the mobile station changes training sequence it uses accordingly.

0066<figref idref="f0013">Figure 12A</figref> shows a pair of transmitters of OSC or co-TCH (co-traffic channel) for uplink transmission. Most of the components are standard and will not be described in detail. Each transmitter includes a training sequence repository 210 a target training sequence set and a new training sequence set generated using one of the methods described above. The two transmitters employ the same carrier frequency and the same timeslot. This is similar to the multi-user signal generated by the network, but in this case, the respective components are generated in respective mobile stations rather than in a single transmitter. The training sequences used for uplink transmission are assigned by the network and will change when a given mobile station is handed off to a different coverage area.

0067<figref idref="f0014">Figure 12B</figref> shows a receiving apparatus composed of two receivers for receiving respective transmissions from the pair of mobile stations of <figref idref="f0013">Figure 12A</figref> in OSC or co-TCH for uplink transmission. Most of the components are standard and will not be described in detail. The receivers include a training sequence repository 100 containing a target training sequence set and a new training sequence set generated using one of the methods described above. One of these, referred to as training sequence A, is used by the top receiver to perform timing estimation, joint channel estimation/detection, and another of these, referred to as training sequence B, is used by the bottom receiver to perform timing estimation, joint channel estimation/detection.

0068In some embodiments, the approaches described herein are used to produce training sequences for the GSM frame format described with reference to <figref idref="f0001">Figure 1</figref>. More generally, the approaches can be applied to transmit frame formats in which the content for a given user comprises at least a respective training sequence and payload, the payload simply comprising any non-training sequence content.

0069In all of the embodiments described, SNR degradation has been used as an optimization criterion for optimizing cross-correlation properties of sequences. More generally, other optimization criterion can be used to optimize the cross-correlation properties of sequences. Specific examples include: <ol id="ol0005"><li>1) parameters related to the amplitude of cross-correlation coefficients (the maximal value, the average value, the variance, etc.)</li><li>2) simulation-based optimization;</li><li>3) other correlation optimization criteria.</li></ol>

0070In some embodiments, each base station is encoded with the entire target training sequence set and the entire new training sequence set. For example, for transmitting purposes, the training sequence repository 200 of a base station, may be configured with the entire target training sequence set and the entire new training sequence set.

0071In another embodiment, each base station, or more generally, each transmitter, is encoded with at least one training sequence from the target training sequence set (for example one training sequence of all of the training sequences), and at least one training sequence from the new training sequence set (for example one training sequence or all of the training sequences). The training sequence(s) from the new training sequence set may include the training sequence(s) from the new training sequence set that are best paired with the training sequence(s) from the target training sequence set. In some embodiments, each base station is configured with such training sequences during network setup.

0072A transmitter or receiver encoded with at least one training sequence from a set consisting of the target set, and at least one training sequence from a set consisting of the new training sequence set may also be encoded with one or more training sequences other than the target training sequence set and the new training sequence set.

0073In some embodiments, each receiver, for example each mobile station, is encoded with at least one (for example one or all) training sequence of the target training sequence set and at least one (for example one or all) of the training sequences of the new training sequence set. For example, for transmitting purposes, the training sequence repository 210 of a mobile station, may be configured with the at least one target training sequence and the at least one new training sequence. When encoded with all of the training sequences of the target training sequence set and the new training sequence set, this will allow the mobile station to perform handoffs between base stations that are assigned any training sequence(s) from the target training sequence set and/or the new training sequence set.

0074A transmitter or a receiver that is encoded with a training sequence is a transmitter or a receiver that has the training sequence somehow stored and useable by the transmitter or receiver. A transmitter or a receiver, such as a base station or a mobile station, having a particular training sequence is a transmitter or a receiver that is able to use particular training sequence. This does not convey an active step of storing the training sequence on the mobile station, although it may be preceded by such an active step. It may have been previously stored for example during device configuration.

0075Numerous modifications and variations of the present disclosure are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims, embodiments may be practiced otherwise than as specifically described herein.