High-Speed Packet Data Access (HSPA)
10.2 HSPA RRM Functions
10.2.1 Channel-Dependent Scheduling for HS-DSCH
The scheduling mechanism controls to which user the shared-channel transmission is directed at a given time instant. In each TTI, the scheduler decides to which user(s) the HS-DSCH should be allocated. The NB also dynamically applies AMC mechanism that defines at what data rate transmission is optimal.
The principle of channel-dependent scheduling is based on fundamental assumption of independent channel conditions for different users at each time instance. Such an assumption is truly satisfied with Raleigh fading. Independence of signal level fading for different users leads to high likelihood that there is a user with the best possible radio channel conditions (at fading peak) at each time instance, as shown in Figure 10.3.
A radio link to this user is likely to be of good quality, therefore, a high data rate channel can be configured for a selected user in that TTI.
The highest overall throughput of the scheduler is achieved in the case of statisti- cally independent fading on radio links to different users. In that way, the scheduling algorithm just exploits Rayleigh fading properties rather that mitigate its impact.
time TTI = 2 ms
UE 2 UE 1 UE 2 UE 2 UE 1 UE 2 UE 2 UE 1
UE 1 UE 2 Signal
level
Figure 10.3 Channel-dependent scheduling.
10.2.2 Rate Control, Dynamic Resource Allocation, Adaptive Modulation and Coding
The data rate control in HSDPA is implemented by dynamically adjusting the channel coding rate as well as dynamically selecting between QPSK and M-QAM modulation.
Higher-order modulation such as QAM allows for higher bandwidth utilization than QPSK, but requires a higher receivedEb/N0. Consequently, M-QAM (16QAM, 32QAM, 64QAM) is mainly useful in good channel conditions often at shorter ranges compared with QPSK. The data rate is selected independently for each 2 ms TTI by the NodeB and the rate control mechanism can therefore track rapid channel variations.
The AMC is sensitive to measurement error and delay. In order to select the appropriate modulation, the scheduler must be aware of the channel quality. Errors in the channel estimate will cause the scheduler to select the wrong modulation/coding scheme and respective transmission power allocated for HS_DSCH. The total power of HSDPA carrier is shared between Release 99 channels, which are power controlled and HS-DSCH, which is rate controlled. The HS-DSCH data rate is then selected to match the radio conditions and the amount of power instantaneously available for HS-DSCH transmission.
The feedback to NodeB for selection of the HS-DSCH transport format is supported in two ways:
1) The UE estimates the downlink channel quality and calculate a suitable transport format that is reported to the NodeB.
2) The NodeB may determine the transport format based on power control gain of the associated dedicated physical channel.
10.2.3 Hybrid-ARQ with Soft Combining, HARQ
HARQ is a link adaptation technique that works along with AMC by reducing the sensitivity to channel measurement and respective prediction error and traffic fluctua- tions. With HARQ, link layer acknowledgements are used for retransmission decisions.
The terminal attempts to decode received transport block, buffers it and reports to the NodeB its success or failure in 5 ms intervals. A relatively short interval allows for rapid retransmissions of erroneous data blocks.
10.2.4 Retransmission Mechanism in the NodeB
The principle of HSPA physical-layer retransmission is shown in Figure 10.4. The HSDPA packet is first received in the buffer in the NodeB. The NodeB keeps the number of packets in the buffer even after sending it to the UE. In case of packet decoding error, the NodeB retransmits buffered packet.
The UE buffers the received data as well and then is able to combine both (or all) received replicas of the packet. As observed, the RNC is not involved in the described retransmission process that is performed at the physical layer.
Should physical-layer retransmission fail, then RNC based retransmission may still be applied on top, as shown in Figure 10.4. Typically, RLC level retransmission is applied in RNC due to signalling errors or in connection to serving cell change in mobility opera- tion. The HSDPA RLC retransmission is normally 100 times less probable than NodeB retransmission.
High-Speed Packet Data Access (HSPA) 177
UE NB RNC
1. Packet Data to NodeB buffer
2. DL transmission 3. NACK
4. Retransmission
Combine
3. RLC ACK
Figure 10.4 Packet retransmission principle in NodeB.
10.2.5 Impact to Protocol Architecture
The transfer of RRM functions, such as fast scheduling and packet retransmission, to the NodeB leads to the changes in protocol architecture. The specific HSDPA additions to user-plane protocol architecture are shown in Figure 10.5. In particular, the new protocol entity MAC-hs (high-speed) is placed in BTS. The MAC-hs handles the scheduling and user packet priority. The RNC retains part of MAC-d (‘d’ stands for dedicated) functionality, specifically, transport channel switching as all other functionalities while scheduling and priority handling are moved to MAC-hs in BTS.
The higher layers, starting from RLC, have no impact from HSPA.
With HSUPA (E-DCH) there is also a new MAC entity added to the BTS, MAC-e, as shown in Figure 10.6.
A new HSUPA related protocol entity is also introduced in the terminal. This is a new MAC entity, MAC-es/e. The MAC-e resides in both the UE and NodeB and is responsible for fast HARQ retransmissions, scheduling and demultiplexing. The
L2
L1 RLC
L2
L1 L1
L2
(transport)
L2
(transport)
L1 HS- DSCH
FP
lur Iub
PHY MAC
PHY RLC
Uu
MAC- hs / MAC-
ehs
HS- DSCH
FP
HS- DSCH
FP
HS- DSCH
FP MAC-c/sh
MAC-D
Figure 10.5 Protocol architecture of HS-DSCH, configuration with MAC-c/sh [2].
PHY PHY
EDCH FP EDCH FP
UE Uu NodeB Iub
DCCH DTCH
TNL TNL
DTCH DCCH
SRNC MAC-d
MAC-e
MAC-d
MAC-es / MAC-e
MAC-es
Iur
TNL TNL
DRNC Figure 10.6 Protocol architecture of E-DCH [3].
MAC-es resides in RNC and the UE and responsible for in-sequence delivery of retransmitted packets (reordering) and enables uplink soft combining for data when UE is involved in inter-NodeB soft handover.
10.2.6 HARQ Schemes
There are many schemes for implementing HARQ: chase combining, rate compatible punctured turbo codes and incremental redundancy [1]. All these methods are based on retransmission of erroneous transport blocks, while some retransmit the same block, the other retransmit erroneous block with addition reductant information (parity bits, different puncturing pattern). The decoder at the receiver combines these multiple copies of the transmitted packet weighted by the received SNR.
The hybrid-ARQ functionality spans both the MAC-hs and the physical layer. As the MAC-hs is located in the NodeB, erroneous transport blocks can be rapidly retransmit- ted. Hybrid-ARQ retransmissions are therefore significantly less costly in terms of delay compared to RLC-based retransmissions that involve RNC. There are two fundamental reasons for this difference:
1) There is no need for signalling between the NodeB and the RNC for the hybrid-ARQ retransmission. Consequently, any Iub/Iur delays are avoided for retransmissions.
Handling retransmission in the NodeB is also beneficial from a pure Iub/Iur capacity perspective; hybrid-ARQ retransmission comes at no cost in terms of transport-network capacity.
2) In order to reduce signalling over Iub/Iur interface, the RLC protocol is configured with a status report once per several TTIs, therefore it is slow compared with HARQ.
In HSDPA, the hybrid-ARQ operates per transport block or, equivalently, per TTI.
That is, whenever the HS-DSCH CRC indicates an error, a retransmission of trans- port block is then requested.
Incremental redundancy is the basic scheme for soft combining; that is, retrans- missions may consist of a different set of coded bits than the original transmission.
Different redundancy versions, that is, different sets of coded bits, are generated as part