TSGR1#18(01)0079
`
`TSG-RAN Working Group 1
`Boston
`January 15-19
`
`Agenda item:
`Source:
`Title:
`Document for:
`
`AH24, HSDPA
`Lucent Technologies
`Variable TTI proposal for HSDPA
`Discussion and decision
`
`1 Introduction
`The notion of using a variable length transmis sion time interval (TTI) for the HS-DSCH was introduced in [1]. The
`benefits of this approach are elaborated upon here along with some simulation results.
`2 Motivation
`Previous approaches ([2] and [3]) have used fixed length TTIs for transmission of HSDPA frames. In [2], a TTI duration
`of 3.33ms (5 slots) has been used, while in [3] the TTI duration was fixed at 0.667 ms (1 slot). Keeping the TTI fixed
`results in difficulties on the following points.
`
`a) Frame Fill Efficiency: When the bit-rate assigned to a user is high, a large TTI could result in there being insufficient
`data to transmit over the duration of one TTI. This results in “frame-fill” inefficiency. As an example if the user is
`assigned a bit-rate of 10.8 Mbps and the TTI is 3.33ms, the HSDPA frame size is 4496 bytes. For typical Internet
`traffic and packet sizes, this could result in considerable inefficiency.
`
`b) Minimum Bit Rate Allowed: With a small TTI and a reasonable value for the minimum code block size, the minimum
`assignable bit-rate may be too high. As an example, consider a minimum code block size of 320 bits (as used in [3])
`and a TTI of 0.667 ms. The resultant minimum assignable bit-rate is 480 Kbps. Furthermore, small code block sizes
`would result in lower Turbo decoding gains and consequently, require higher energy-per-bit for same error rate.
`
`c) MCS Level for Retransmissions: The use of fixed TTI makes it necessary to use the same MCS level for
`retransmission (if soft combining is to be done) as the one used for the first transmission of a frame. The channel
`conditions, available power and/or code space at the time of retransmission (within the same cell or selected cell)
`may not permit the use of the same MCS level as the original transmission. By making the TTI variable, incremental
`redundancy (IR) operation can be made adaptive wherein retransmissions can be at a different MCS level from the
`original transmission (see [4] and [5]).
`
`d) Signalling Overhead: The user identification overhead with fixed and small TTI is higher in the low to medium data
`rates range as compared to a variable TTI. This is due to the fact that in a scheme with fixed TTI, lower rate implies
`smaller code block sizes. Consequently the user identification and other HARQ control overhead per information bit
`is high. The variable TTI approach allows the sub-block transmission over a larger number of slots for low data
`rates. This reduces the user identification and other HARQ control overhead per sub-block transmission.
`
`e) Flexibility: The code block size is coupled to the data rate with fixed TTI. With variable TTI, different rates can be
`achieved for the same code block size by varying the TTI. The variable TTI proposal also allows for different code
`block sizes at the same MCS level. This will make sure that the appropriate code block size is chosen for a given data
`rate and a given user buffer size. This achieves a good tradeoff between signalling overhead and padding. For
`example, suppose the supportable rate is 960Kbps and there are 1280 bits (4 transport blocks of size 320 bits each)
`in the user buffer. With a single slot fixed TTI these bits will be transmitted as two code blocks over two slots.
`Therefore, overheads (e.g. CRC) will be associated with each of the two transmissions. With a fixed TTI of five slots,
`a total of 3200 bits will have to be transmitted resulting in a padding of 1780 bits. With the variable TTI approach as
`proposed here, 1280 bits can be transmitted as a single code block over 2 slots. This results in only one set of
`overheads with no padding and thus provides more efficient transport as compared to either of the fixed TTI
`options.
`
`1
`
`APPLE 1013
`
`

`

`In this contribution we illustrate how the use of a variable length TTI with the granularity of 0.667 ms (1 slot) coupled
`with fixed length code blocks overcomes some of the difficulties with the other approaches and allows for more efficient
`operation of the HS-DSCH. The concept is elaborated upon in Section 2. Section 3 contains simulation results followed
`by concluding remarks in Section 4. A text proposal based on the variable length TTI concept is outlined in Section 5.
`
`3 Dynamically varying TTI
`In a scheme with fixed TTI, the code block size is determined uniquely by the MCS level and therefore, the number of
`information bits in the TTI changes with the data rate. With the variable TTI approach, the duration of the transmission
`is varied while the code block size in bits is kept fixed.
`
`The data rates for the variable TTI scheme are given in Table 1 [5]. Note that for a given data rate, up to four different
`code block sizes can be chosen depending upon the data backlog in the user buffer. The flexibility of using different
`code block sizes for the same data rate avoids huge performance loss due to frame-fill inefficiency and would also limit
`signalling overhead as discussed in Section 1.
`
`Table 1. Data Rates [assumes 20 channelization codes at SF=32]
`Transmission Time Interval (TTI) [number of
`slots]
`
`MCS
`
`Data
`Rate
`
`[Kb/s]
`
`Modulation
`
`Effective
`coding
`rate
`
`16
`Transpor
`t blocks
`per TTI
`
`[code
`block =
`5120
`bits]
`
`8
`Transpor
`t blocks
`per TTI
`
`[code
`block =
`2560
`bits]
`
`4
`Transpor
`t blocks
`per TTI
`
`[code
`block =
`1280
`bits]
`
`2
`Transport
`blocks
`per TTI
`
`[code
`block =
`640 bits]
`
`16
`
`8
`
`4
`
`2
`
`1
`
`16
`
`8
`
`4
`
`2
`
`1
`
`16
`
`8
`
`4
`
`2
`
`1
`
`16
`
`8
`
`4
`
`2
`
`1
`
`[actual
`coding +
`repetition]
`
`0.0125
`
`0.0250
`
`0.0500
`
`0.1000
`
`0.2000
`
`0.4000
`
`0.8000
`
`0.8000
`
`1
`
`2
`
`3
`
`4
`
`5
`
`6
`
`7
`
`8
`
`60
`
`120
`
`240
`
`480
`
`960
`
`1920
`
`3840
`
`7680
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`QPSK
`
`16-QAM
`
`4 Simulation Results
`With a single slot fixed TTI, the minimum code block size will have to be small to keep the minimum supported data rate
`reasonable. However, if the TTI is allowed to be variable, the minimum code block size can be picked to be larger, while
`keeping the minimum supported data rate reasonable. The result will be improved Turbo coding performance at the
`expense of multi-slot transmission at the lower rates. Simulation results were conducted to compare the frame error rate
`as a function of block size while keeping the data rate fixed.
`
`2
`
`

`

`The simulation parameters are given in Table 2.
`
`Table 2. Simulation parameters
`
`Parameter
`No of iterations for Turbo Codes
`Metric for Turbo Code
`Turbo Code Rates
`Input to Turbo Decoder
`Turbo Interleaver
`
`Value
`8
`Max
`0.2-0.8
`Soft
`As per 3GPP (modified to handle large code
`blocks)
`
`The Frame Error Rate (FER) as a function of the received ˆ
`/or
`I
`I
`is shown in Figure 1. The results are given for five
`oc
`different code block sizes i.e., 320, 640, 1280, 2560 and 5120 bits and two different modulation and coding schemes i.e.,
`MCS 4 (480Kb/s) and MCS5 (960 Kb/s). For MCS4 at 1% FER, 320 bits code block needs 1.6dB higher Eb/No compared to
`5120 bits code block size.
`
`1
`
`0.1
`
`0.01
`
`FER
`
`0.001
`
`-8
`
`5120 MCS4
`2560 MCS4
`1280 MCS4
`640 MCS4
`320 MCS4
`5120 MCS5
`2560 MCS5
`1280 MCS5
`640 MCS5
`
`-6
`
`-4
`I^or/Ioc [dB]
`
`-2
`
`0
`
`Figure 1. FER for different code block sizes.
`5 Concluding Remarks
`We have presented a variable TTI concept where the code block size in bits is fixed and the transmission duration is
`varied in order to achieve different data rates. The variable TTI concept allows using larger code block sizes even for
`lower data rates in order to get maximum Turbo coding gains. For higher data rates, the transmission time is kept to
`minimum to fully exploit the scheduling gains, while still achieving high Turbo interleaving gains. Another aspect of our
`design is that retransmitted blocks can be at a different MCS level as compared to the original transmission. The variable
`TTI scheme also presents minimum user identification overhead (particularly at lower data rates) because the large
`number of information bits (code block) can be transmitted over larger number of slots. Note that the user identification
`overhead is per code block and is higher for smaller code blocks.
`
`3
`
`

`

`6 Text Proposal for TR
`Based on the arguments and results presented in this document, variable TTI offers some advantages over a fixed TTI
`approach. As such it should be considered as part of the feasibility study. Therefore, the following text is recommended
`to be included in Section 6.1.2, “HSDPA Physical Layer Structure in the Time Domain” of 3G TR25.848.
`
`---------------------------------------------------------------------------------------------------------------------------------------------------
`
`Variable TTI schemes wherein the code block size stays fixed but the duration of transmission to the user is changed
`based on the chosen rate, afford several benefits.
`
` The minimum code block size can be large without the minimum assignable bit-rate being too large. Reasonable
`code block sizes help in fully exploiting Turbo interleaving gains.
`
` With a minimum TTI of one slot, better frame-fill efficiency is obtained for bursty traffic at high data rates.
`
` The MCS level for retransmission of a frame can be different from the MCS level for the original transmission. This
`is very useful in cases when the channel conditions, available power and code space at the time of retransmission
`are different from that of the original transmission. This is also true with fast cell site selection (FCSS) when the
`selected cell does not have the same available resources for the HS-DSCH as the original cell.
`
` By allowing for multiple code block sizes for a given MCS level, the chosen code block size for a given MCS level
`(and thereby the transmission duration) can be better matched to the data backlog in the user buffer.
`
` Signalling overheads, such as those required for user identification and Hybrid ARQ operation are per code
`block. Since the minimum code block size can be designed to be large, the overhead-to-payload ratio can be
`lower with the variable TTI approach.
`
`Variable TTI schemes with single-slot granularity should be compared to schemes that use fixed TTIs in terms of
`performance, flexibility and complexity.
`
`---------------------------------------------------------------------------------------------------------------------------------------------------
`
`7 References:
`[1] Lucent Technologies, “Downlink and Uplink Channel Structures for HSDPA,” TSGR1#17(00) 1381, Stockholm,
`Sweden, November 2000.
`
`[2] Motorola, “HSDPA System Performance Based on Simulations (II)”, TSGR1#17(00) 1397, Stockholm, Sweden,
`November 2000.
`
`[3] Ericsson, “Performance Comparison of Chase Combining and Incremental Redundancy for HSDPA,” TSGR1#17(00)
`1428, Stockholm, Sweden, November 2000.
`
`[4] “Asynchronous and Adaptive Incremental Redundancy (A 2IR)”, Lucent Technologies, TSGR1#17(00) 1382,
`Stockholm, Sweden, November 2000.
`
`[5] “A2IR – An asynchronous and adaptive HARQ scheme for HSDPA”, Lucent Technologies, TSGR1#18(01) 0080,
`Boston, USA, January, 2001.
`
`4
`
`