High Speed Downlink Packet Access - HSDPA

HSDPA stands for “High Speed Downlink Packet Access”. As the name suggests, this is a piece of UMTS functionality designed to deliver downlink packet data at very high data rates. It is a release 5 feature. It achieves its aim by using the following techniques:

• Use of shared channel concept Rather than constantly allocating and deallocating dedicated channels to individual users, users share a high bandwidth channel – the HS-DSCH (High Speed Downlink Shared Channel). This allows the system to operate with a “fat pipe”. An analogous situation in queuing theory is that an M/M/1 queuing system is more efficient than an M/M/n system.

·         Use of Hybrid ARQ/Incremental Redundancy

·         Use of multicode transmission

·         Availability of high level modulation (QAM)

·         Link Adaptation

·         Fast scheduling, facilitated by a short TTI.

HYBRID ARQ/INCREMENTAL REDUNDANCY

ARQ (“Automatic Repeat Request”) schemes are ones in which, when a data unit arrives, a check is made on the integrity of this data. If the check fails, a retransmission is invoked. The integrity check is normally made by means of a CRC. If the data unit (usually including the CRC) is additionally protected by means of an error correcting code, the scheme is referred to as a “Hybrid ARQ” scheme. HARQ schemes are further subdivided:

TYPE I HARQ   

In these schemes, an erroneous data block is discarded, and re-transmissions are treated independently. There is no attempt to combine re-transmitted blocks with old attempts. Re-transmitted blocks are identical to the original block.

TYPE II HARQ 

An erroneous block is not discarded, but is combined with subsequent re-transmissions. Re-transmitted blocks can be constructed with different coding rates to the original blocks. The re-transmitted version of the block may or may not be self-decodable – for example, it may have a very low proportion of systematic bits.

TYPE III HARQ

This is identical to type II HARQ, except that each re-transmission is self ecodable.

HSDPA – GENERAL PRINCIPLES

There are a number of channels involved in supporting a service to a given UE using HSDPA. These are:

HS-DSCH

High Speed Downlink Shared Channel. This is the high speed “fat pipe” on which the packet data will be transmitted. The HS-DSCH may be carried on several physical channels (HS-PDSCH)

HS-SCCH

High Speed Shared Control Channel. This is a physical channel used to carry signalling (principally HARQ and transport format information) related to the HS-DSCH. A cell may have several HS-SCCHs.

HS-DPCCH

High speed DPCCH. This is a special DPCCH assigned to the UE which carries CQI and HARQ ACK/NACK signalling in the uplink.

ASSOCIATED DEDICATED CHANNEL

The HS-DPCCH can only exist together with an uplink DPCCH. The uplink associated channel DPDCH, DPCCH and the HS-DPCCH are I/Q code multiplexed. The HS-DPCCH and DPCCH are both transmitted on the Q channel (using separate channelisation codes). The DPDCH part of the uplink associated channel can also be used to transmit data.

One function of the downlink associated dedicated channel is to provide power control for the uplink channels, including the uplink HS-DPCCH, but it can also be used to transmit downlink data as well as partake in soft handover processes.

HARQ PROCESSES

In outline, the UE monitors the HS-SCCHs looking for an indication that there is about to be some data destined for it. When there is data indicated, the HS-SCCH contains enough information to enable the UE to decode the data on the HS-DSCH. The timing of the 2 channels is arranged so that the signalling can be decoded in time to allow decoding of the user data.

When the user data block is received on the HS-DSCH, it is checked to see if it has been received correctly. If yes, an ACK (positive acknowledgement) is sent on the uplink HS-DPCCH. If not, it is stored at the receiver to allow combination with a repeat attempt, and a NACK (negative acknowledgement) is sent on the uplink HS-DPCCH. The function attempting to correctly receive a particular data block is referred to as a HARQ process.

The system may have several HARQ processes running concurrently for a given UE – see the following figure for an example.

Here we see 3 HARQ processes. HARQ process 0 is assigned to receiving block number 23. The transmission fails on the first attempt, the received data is stored and a NACK is generated. It then gets retransmitted, and combined with the first attempt. This is decoded successfully and an ACK is generated. Process 0 is then free for reassignment to another block. Process 1 is assigned to block 24, which fails on the first attempt and is retransmitted. Process 2 is assigned to block 25, which is successful on the first attempt, then is reassigned to block 26.

ACK/NACK TIMING

The timing relationship between the HS-DSCH and the HS-DPCCH is complex and is described in detail in 3GPP 25.211v5.5.0. For the purposes of the simulation study here, we consider the HS-DPCCH block containing the ack/nack to a given HS-DSCH block to start 10.5 timeslots after the start of that block. The following sketch illustrates this - the small tick marks are timeslots and the large ones are HS-DSCH TTI boundaries.

REDUNDANCY VERSIONS  

On each retransmission of a block, the set of bits that are actually transmitted may be changed – for example, the set of parity bits chosen for transmission on the second attempt may be different to the set chosen on the first attempt. These different versions of the coding format for the data block are called “redundancy versions”. In order to attempt a decoding, the receiver needs to know the redundancy version used when each transmission attempt is made (in order to perform the HARQ combining process). This information is transmitted on the HS-SCCH. The timing of the HS-SCCH transmissions relative to the HS-DSCH is therefore important.

An additional parameter which can be varied between retransmissions of a block is the bit-to-constellation mapping (sometimes referred to as “Constellation Re-arrangement” – CoRe) in the case of 16-QAM modulation.

16-QAM

In order to achieve the high data rates needed for HSDPA, the option of using 16-QAM modulation has been introduced. QAM is a standard digital modulation technique, details of which can be found in many textbooks on digital communication theory.

QAM CONSTELLATION

The diagram above shows the constellation used for 16-QAM in HSDPA. Bits are taken in groups of 4 and used to obtain the appropriate QAM symbol. Before the symbol lookup, a bit –rearrangement is performed. There are 4 possible re-arrangements as shown in the following table:

 Note that b=0 corresponds to the trivial case of no re-arrangement (straight copy). The bit pair i1q1 indicates which quadrant we’re in, and the bit pair i2q2 indicates which of the 4 symbols within the quadrant to select. The reason for providing the re-arrangement is that, to take an example, if a modulation symbol near to 0011 is received, the receiver’s confidence in making a judgement that the symbol transmitted was in the first quadrant is higher than its confidence in making a judgement about exactly which symbol in the first quadrant was actually transmitted. In other words its confidence about i1q1 is higher than its confidence about i2q2. Bit re-arrangement gives the system the chance to transmit other bits in the high confidence positions.

When doing this, in the LLR formula, x0 corresponds to the set of QAM symbols in the yellow half, and x1 to the set of QAM symbols in the green half. We are computing distances in the real (I) component. Note that we don’t need to compute the I distances to all symbols in a column because they are all the same. This technique for an efficient implementation of a QAM slicer is described in ITU Study Group 15 tdoc CF-038 (Jan 2001).

HS-DSCH PHYSICAL CHANNEL FORMAT

The HS-DSCH is carried on multiple physical channels. Each physical channel has a fixed SF equal to 16. Each channel is also based on a 3 timeslot (2mS duration) radio frame (a radical difference between HSDPA and release 99 channels!). The HS-DSCH timeslot structure consists entirely of user data – there are no pilot/TPC/TFCI fields.

The modulation type is either QPSK or 16-QAM.

HS-DSCH CHANNEL CODING

For channel coding, the HS-DSCH uses the release 99 turbo coder.

HS-DSCH Rate Matching

The rate matching used on the HS-DSCH is based on the release 99 rate matching algorithm. However, an architecture based on 2 separate stages of rate matching has been adopted in order to facilitate the incremental redundancy operation. The overall rate matching architecture is illustrated in the following diagram:

Conceptually, each HARQ process has its own Virtual Incremental Redundancy (VIR) buffer. The size of the VIR buffer is one of the channel parameters which consitute the current HS-DSCH format. The received softbits are stored in the VIR buffer and retained for possible subsequent repeat transmissions. Each repeat transmission may produce some softbits in positions in the VIR buffer already containing softbits from a previous transmission attempt. In this case, the new softbits are simply added to the existing soft values.

The function of the first rate matching stage is to match the data rate of the incoming stream to the capacity of the VIR buffer. In this stage, the systematic bits are not touched. The second rate matching stage matches the capacity of the VIR buffer to that of the physical channels assigned to the HS-DSCH.

Illustration of Rate Matching and Transport Channel Parameters

We consider the following concrete example to illustrate the parameters involved in specifying a HS-DSCH format:

Param		Value		Comment
Inter-TTI Distance		3		(1)
Info. payload		4704
Number of HARQ Processes		2		(2)
Number of SMLs per HARQ process		9600
Number of physical channel codes		4
Modulation type		QAM
CRC length 24		24		(3)

Comments:

(1) Each TTI is transmitted in a 3TS HS-DSCH radio frame, of duration 2mS. This example is set up with an inter-TTI distance of 3, meaning that only every 3rd HS-DSCH radio frame is available for transmission.

(2) The number of HARQ processes that can be simultaneously active is a channel parameter. If 2 HARQ processes are active (i.e. 2 blocks are in the process of being received), and a new block arrives for transmission, it is buffered at the transmitter until one of the existing HARQ processes terminates i.e. either the block is correctly received, or the maximum number of transmission attempts is reached.

(3) This is the same for all HS-DSCH transmissions.

(4704 information bits + 24bit CRC) x 3 = 14172 bits                                                            
Add 12 tail bits for turbo decoder = 14184 bits.                                                                                      

The VIR size for a given HARQ process is 9600 bits, therefore for first RM stage,

Stage1_deltaNTTI = 9600-14184 = -4596

So, 4596 bits must be punctured by the first stage of rate matching.

On each physical channel, each HS-DSCH radio frame has 3840000x0.002 = 7680 chips. Sincethe SF is fixed at 16, this corresponds to 7680/16 = 480 QAM symbols. Each QAM symbol encodes 4 bits, so this corresponds to 1920 bits.

Given we have specified 4 physical channels, there are 1920x4 = 7680 bits available. The second stage of rate matching therefore has to puncture 9600-7680 = 1920 bits, so

Stage2_deltaNTTI = -1920

Redundancy Version       

The actual sequence of bits to be punctured in the second stage of rate matching depends upon a pair of parameters called s and r. These are obtained at the UE from information received from the HS-SCCH. Once s and r have been fixed, the rate matching parameters eini, eplus and eminus for the second RM stage can be determined. S and r are called the redundancy version parameters and are encoded, together with the constellation version parameter b, in an integer called Xrv. The relationship between the values of Xrv and s, r and b combinations is given in the following tables:

RV Parameters for QAM

Xrv	s	r	b
0	1	0	0
1	0	0	0
2	1	1	1
3	0	1	1
4	1	0	1
5	1	0	2
6	1	0	3
7	1	1	0

RV Parameters for QPSK

Xrv	s	r
0	1	0
1	0	0
2	1	1
3	0	1
4	1	2
5	0	2
6	1	3
7	0	3

STRUCTURE OF HS-SCCH

The HS-SCCH also is built on a 3 TS-per-frame structure. The information in an HS-SCCH frame is split into 2 parts. Part 1 is transmitted entirely in the first slot, and part 2 in the subsequent 2 slots.

The notation in this diagram is:

xccs	7 bit channelisation code set info
xms	1bit modulation scheme
xtbs	6bit transport block size info
xhp	3bit HARQ process info
xrv	3bit redundancy version info
xnd	1bit new data indicator
X1	8bits input to rate 1/3 convolutional coder
Z1	48 bits output (including bits for trellis termination)
R1	40 bits after puncturing

xccs 7 bit channelisation code set info    
xms 1bit modulation scheme

xtbs 6bit transport block size info
xhap 3bit HARQ process info     
xrv 3bit redundancy version info 
xnd 1bit new data indicator         
X1 8bits input to rate 1/3 convolutional coder    
Z1 48 bits output (including bits for trellis termination) 
R1 40 bits after puncturing

The 16 bit UE Id (H-RNTI) is used to produce 40 masking bits as follows: rate 1 convolutionally encode (producing 48 bits), then puncture 8 bits. These 40 bits are then x-or'd with the 40 R1 bits.

The CRC is computed from the concatenation of the 8 bits of part 1 and the 13 bits of part 2. It is then masked with the 16 bit UE ID before concatenation with the 13 bits of part 2.

The physical channel supporting the HS-SCCH is a fixed SF (=128) channel containing just data bits (i.e. no pilot, TFCI, TPC etc). Several such HS-SCCHs may exist in a cell. If the UE has not detected an HS-SCCH destined for it in the previous sub frame, it must monitor all HS-SCCHs, since a message destined for it may appear on any of them in principle. If the UE did receive a message destined for it in the previous sub frame, it need only look at the HS-SCCH on which this message was received since the network will continue to use this one while there is HS-DSCH data in the process of transmission.

Part 1 of the HS-SCCH will be scrambled by a UE-specific (i.e. based on the UE id) scrambling code. The idea is that the UE can make a preliminary decision about whether a given monitored HS-SCCH frame is for itself by looking at part 1. Only if it passes this test does the UE go ahead and decode part 2. There are therefore two performance parameters of interest here:

Prob. of part 1 detection fail pf.  

If part 1 detection fails, a transmission which is actually for the UE will be missed since the UE rejects it. The higher pf is, the lower is the throughput on the HS-DSCH.

Prob of part 1 false alarm pfa      

This refers to the case where the UE incorrectly decides that a transmission is for it when in fact it is destined for a different UE or where there was no transmission at all.

(Actually, these are 2 distinct cases). The higher pfa is, the higher is the UEs current consumption (since irrelevant frames are passed through to part 2 decoding).

Now, part 1 does not contain a CRC, so some other criterion is required for the UEs decision over whether the frame is correct or not. A couple of alternatives are:

1)      Yamamoto-Itoh Algorithm    

We consider a state in the trellis to be high confidence if the difference in the path metrics for paths to that state is greater than a given threshold. As the decoding progresses, if a state is deemed “low confidence”, then the surviving path through is labelled “bad”. Once labelled bad, this path always stays labelled bad. At the end, if the chosen path is bad, the decoding is deemed a failure and conversely

2) Final Path Metric Difference Algorithm

Similar to YI, but just compare the difference in the path metrics for the paths merging at the last stage. Again, the decoding is good if this difference exceeds a threshold. If the decoding proceeds to part 2, the normal procedures apply, and the CRC is used to determine if the frame was good or not. Thus, even the false alarms will usually be rejected at part 2.

LINK ADAPTATION

The UE reports the current CQI (Channel Quality Indicator) value to the network. A CQI value is effectively a specification of a transport block size for which the UE would be able to receive at a BLER of <= 10%. Along with the transport block size is specified the number of HS-DSCH physical channel codes and the modulation type.

CQI Value	TB Size	No. of HS-PDSCH	Mod. Scheme	Ref. Power Adjustment (Delta)
0	N/A
1	137	1	QPSK	0
2	173	1	QPSK	0
3	233	1	QPSK	0
4	317	1	QPSK	0
5	377	1	QPSK	0
6	461	1	QPSK	0
7	650	2	QPSK	0
8	792	2	QPSK	0
9	931	2	QPSK	0
10	1262	3	QPSK	0
11	1483	3	QPSK	0
12	1742	3	QPSK	0
13	2279	4	QPSK	0
14	2583	4	QPSK	0
15	3319	5	QPSK	0
16	3565	5	16-QAM	0
17	4189	5	16-QAM	0
18	4664	5	16-QAM	0
19	5287	5	16-QAM	0
20	5887	5	16-QAM	0
21	6554	5	16-QAM	0
22	7168	5	16-QAM	0
23	7168	5	16-QAM	-1
24	7168	5	16-QAM	-2
25	7168	5	16-QAM	-3
26	7168	5	16-QAM	-4
27	7168	5	16-QAM	-5
28	7168	5	16-QAM	-6
29	7168	5	16-QAM	-7
30	7168	5	16-QAM	-8

There are some points to note:

1) This table is valid for UE categories 1-6. The “UE category” is a specification of its HSDPA capability. There are different CQI tables for other UE category values.

2) When the UE is testing, for example, CQI=24, it is asking the question “could I receive a length 7168 transport block tranmsitted with 16-QAM over 5 physical channel codes with a BLER (on first transmission, not including repeats!) <=10% if the transmit power was
PCPICH + Gamma + Delta

where:
PCPICH is the power of the CPICH (primary or secondary depending on where the UE is getting its phase reference from) Gamma is the measurement power offset (signalled from higher layers) Delta is the reference power offset from the table

3) The special value of CQI = 0 means that the UE cannot receive any HS-DSCH transmission within the required criteria.

On the basis of the reported CQI, the network decides the HS-DSCH transport format to use in subsequent transmissions.

HSDPA IMPLEMENTATION IN PLUS-D

PLUS-D has been modified to include some aspects of HSDPA operation. The following points apply:

• To accommodate the 3slot HS-DSCH radio frame as well as the 15 slot DPDCH frame, the fundamental operation has been changed from a frame-based loop to a slot-based loop.

• The HS-DSCH is modelled. Multi-code operation, QPSK and 16-QAM modulation, and the 2 stage rate matching algorithms with HARQ are all implemented.

• The current version of PLUS-D uses a straightforward RAKE receiver. This is probably not the most appropriate receiver architecture. Different approaches may be attempted in future versions of the software.

• The HS-SCCH functionality is not yet available – this has been bypassed for the moment by assuming the UE always has perfectly accurate knowledge of the current transport format on the HS-DSCH.

• The uplink link adaptation (CQI) signalling is not yet modelled.

• The HARQ processing for a given HARQ process is based on the state transition diagram (see next page).

The state “IDLE” applies when the HARQ process is inactive. When the process is chosen to transmit new user data, the state for this process transitions to “NEW”. State “WAIT” is used when data has been transmitted and a reply is pending. On entering “NACKED” state, a timer is started (to reflect the delay caused by the HS-DSCH -> HS-DPCCH timing offset), when this expires, the state transitions to NACKEDRDY, and the frame can be retransmitted.

CONCLUSION

HSDPA is rapidly becoming a commercial reality in numerous networks around the world. Measurements made in live networks show impressive performance. The trengths of Ericsson’s WCDMA radio network design.

•  Positively influence end-user perception of HSDPA performance; and

• Minimize the need for operators to deploy additional cell carriers in order to support new HSDPA services. Following a strong start, HSDPA will continue to evolve through numerous performance enhancements that will both make HSDPA services more appealing to end users and provide the system capacity that is needed to support rapid HSDPA service uptake.