Вы находитесь на странице: 1из 3

Understanding HSDPA’s

implementation challenges
By David Maidment that cell. The changes that support for bursty IP traffic for at the edge of the cell. For a
Product Manager HSDPA enables include better the mobile user. large cell with a diverse range of
picoChip Designs Ltd quality and more reliable, more Instead of sending data us- users, the peak aggregate data
E-mail: david.maidment robust data services. In other ing individual DCHs, HSDPA rate will be in the range of 1-
@picochip.com words, while realistic data extends the downlink shared 1.5Mbps. This can increase to
rates may only be a few Mbps, channel (DSCH), allowing more than 6Mbps as the cell
Since its introduction, 3G tech- the actual quality and number packets destined for many users size decreases to the microcell
nology has been heralded for of users achieved will improve to be shared on one higher- level and beyond. In principle,
its ability to deliver more voice significantly. bandwidth channel called the a picocell could see data rates of
channels and higher-bandwidth high-speed DSCH (HS-DSCH). 8Mbps or more.
pipes. But operators have Burst problems As with wired networks such as To achieve higher raw data
started to realize that while 3G IP is a bursty protocol that de- Ethernet, this allows for the rates, HSDPA uses higher-level
allows for high-quality voice and mands changes to the wideband- more efficient use of available modulation schemes such as
media streaming services, it is CDMA (W-CDMA) protocol bandwidth. On top of that, a 16QAM at the PHY layer, to-
a poor fit for high-speed data. stack to support IP efficiently. faster channel configuration gether with an adaptive coding
High-speed downlink pa- Bursty protocols are a poor process allows the base station scheme based on turbo codes.
cket access (HSDPA) techno- fit with dedicated channels to control the channel more ef- Note that the modulation
logy promises to bridge the gap (DCHs) that are used in existing fectively (Figure 1). scheme is adaptive and changed
between 3G and the Internet, W-CDMA networks. Although on a per-user basis. The spread-
providing an overlay for the ex- the DCH can support many dif- Too many options ing factor used for the HS-
isting protocol stack that en- ferent types of traffic, use of the There are many options avail- DSCH remains fixed at 16, but
ables the delivery of high-speed channel for bursty traffic is quite able to base-station designers the coding rate can vary on a
data to many users in a cell. In- low. This is because the process and operators when dealing per-user basis between 1/4 and
stead of limiting high-speed of channel reconfiguration that with HSDPA. This complicates 3/4. In theory, the protocol al-
data access to fewer than five can be used to tune the DCH for the provision of HSDPA as the lows an uncoded link of 4/4,
users in a cell, HSDPA can de- a change in traffic mix traffic is network is upgraded, but intel- but that is only useful for lab
liver 384Kbps data to up to 30 slow, taking on the order of ligent choices over base-station tests to achieve the theoretical
users. 500ms. implementation can result in maximum of 14Mbps using
These issues have been ad- higher throughput for high- 16-QAM modulation.
Not simple dressed in Release 5 of the 3G revenue services, improving However, many of the re-
But HSPDA is not a simple soft- Partnership Project (3GPP) operators’ margins. sults announced so far fall short
ware upgrade to 3G systems. In standards, which radically The maximum bandwidth of what is expected from a ma-
many respects, the change from changes the network to make it that can be achieved with ture, robust system. In small
Release 99 to HSDPA is as sig- better suited to data traffic. HSDPA depends largely on cell cells (the 3G equivalent of hot
nificant as that from voice-only Support for IPv6 has been in- size. To limit the power needed spots), ver y high data rates
GSM to EDGE—changing both corporated into the core net- to send each bit of information, should be realistically attained.
modulation and the way packets work with a key enhancement the maximum achievable bit Under poor reception condi-
are processed. to provide high-bandwidth rate tends to fall away for users tions, the modulation can vary
There are parts of the
HSDPA standard that are rela-
tively simple to implement us-
ing existing hardware. But, HS-PDSCH High-speed physical downlink shared channel
taken as a whole, HSDPA will "The data bearer"
simply break many deployed Downlink
HS-PDSCH
architectures and will require HS-PDSCH Up to 15, always associated with a DCH
Data bearer: Peak data rate 14.4Mbps
new hardware. Most base sta- HS-PDSCH
SF = 16
tions (also known as Node Bs)
will need significant upgrades
to cope with the increased data High-speed shared control channel
throughput and consequences "The signaling channel"
HS-SCCH
of moving to a more complex
Carries HARQ information and format parameters
protocol. Up to 4 logical channels per UE
HSDPA increases the down- SF = 128
link data rate within a cell
to a theoretical maximum of Uplink HS-DPCCH
14Mbps, with 2Mbps on the HS-DPCCH High-speed dedicated physical control channel
uplink. However, it is not about HS-DPCCH "Packet flow control signaling"
delivering Ethernet bandwidth
Carries HARQ, channel quality information
to one fortunate user. What is HS-DPCCH SF = 256
important is the ability to reli-
ably deliver many sessions of
high-speed, bursty data to a Figure 1: Instead of sending data using individual DCHs, HSDPA extends the DSCH, allowing packets destined for
large number of users within many users to be shared on one higher-bandwidth channel called the HS-DSCH.
as well, possibly reverting to significantly. These changes packets can be retransmitted cremental redundancy can be
QPSK from the higher-order will be difficult to achieve in a quickly at the request of the used. In this scheme, parity
modulation of 16-QAM. Link hardware design that was not mobile terminal. In existing W- bits are sent to allow the mo-
adaptation ensures the highest architected to support them. CDMA networks, these re- bile terminal to combine the
possible data rate is achieved Indeed, some early demon- quests are processed by the information from the first
both for users with good signal strations only implemented a RNC. As with fast scheduling, transmission with subsequent
quality who are typically close few of these features or only better responsiveness is pro- retransmissions.
to the base station and for more achieved limited data rates. vided by HSDPA by processing The scheduler and retrans-
distant users at the cell edge While 16-QAM modulation is the request in the base station. mission manager require large
who may receive data with a the most obvious change and The hybrid automatic repeat buffers to hold all the packets
lower coding rate. The link ad- easy to demonstrate at a trade request (HARQ) protocol de- that might need to be re-sent.
aptation is performed on each fair, the capabilities of the veloped for HSDPA allows effi- This function was not present
transmission timing interval MAC-hs and adaptive control cient retransmission of drop- in earlier functions, so the sup-
(TTI), with the user equipment loops are more important, but ped or corrupted packets. The porting hardware needs to have
sending an estimate of the less visible. Developing, testing protocol was designed to allow been designed in readiness for
channel quality to Node B that and “hardening” these algo- the average delivered band- it for existing implementations
is then used to select the modu- rithms for field deployment are width of HSDPA to be higher to support HSDPA at suffi-
lation and coding rate for that the major challenges for manu- than would be possible if more ciently high data rates.
user on the next transmission. facturers. extensive forward error correc-
tion were to be used. However, Factors impacting scheduling
Moving MAC control Increasing complexity it puts significant demands on It is simple to devise a schedu-
The more important change The improved performance and the base station, since support ling algorithm that will work
that HSDPA makes is to move processor power provided by for HARQ calls for low latency. well for a few users in the labo-
control of the medium access HSDPA implies an increase in The latency demanded for effi- ratory with artificially gene-
control (MAC) layer from the complexity. Many high-speed cient HARQ support calls for rated constructive fading condi-
radio network controller (RNC) feedback loops are needed to retransmissions to be pro- tions, but it is much harder to
into the base station. Crucially, implement HSDPA efficiently cessed within 2-7ms. But the develop one that works robustly
this move enables the use of fast- and provide users with the best feedback loop that allows in the field for many users—all
scheduling algorithms where, data rates possible. HARQ to be implemented is not with different, complicated and
under constructive fading con-
ditions, users are served data
based on channel-quality esti-
mates. This compares to risking R4 Tx symbol rate R4 Tx chip rate Tx
high error rates to be experi- sample
HS Tx rate
enced by users in poor reception Transport lub/FP encode
HS Tx chip rate
To/from
conditions using a conventional layer
termination
MAC-hs ADC/DAC
Channel element controller
user-priority or round-robin lub termination and RF
scheme, where the scheduler HS-DPCCH decode Rx
R4/HS-DPCCH
uses average channel conditions Rx chip rate
sample
R4 Rx symbol rate rate
to select the modulation and
coding scheme used. As a result, Device 1 (PC 102) Device 0 (PC 102)
fast scheduling works hand-in-
hand with the algorithms used PowerPC SDRAM SRAM
to select optimum modulation
and coding schemes. Radio network layer
Transport network layer
This increases the respon- - NBAP (TS5.433);
- NBAP transport (ATM, AAL5)
siveness of the base station. The - Transport signaling (ALCAP) - L1 resource management and
16-QAM coding change in- - Data stream transport (ATM, AAL2) measurement
- Specific O, A and M - Framing protocol termination for
creases the peak speed in the dedicated and common channels
same way that a high-powered
engine can boost the perfor-
mance of a car, but it is the Figure 2: Protocols such as HSDPA work well on parallel-processing architectures.
MAC change that makes
HSDPA deliver a real-world For example, the TTI used one that exists in Release 99 changing situations. Many cir-
speed increase, much like re- for modulation and coding se- base stations, as that function cumstances will affect real-
placing a learner driver with a lection for individual frames in sits in the RNC for existing world systems, not the least of
Formula One racing driver. HS-DSCH is just 2ms, com- DCH and DSCH transmissions. which are evolving capabilities
Performance will be notice- pared with a typical time of Hence, not only must things of the terminals themselves,
ably better even if 16-QAM 10ms (and up to 80ms) for the work faster, but many functions whether they are handsets or
modulation cannot be used. TTI used for power control in are new, adding to the capabili- data cards inserted into PCs.
This demonstrates how a shift the existing Release 99 shared ties and intelligence of Node B. Latency demands of HSDPA
in the 3G architecture from a channel. Furthermore, the al- In addition to fast retrans- mean that designs will react dif-
traditional “dumb pipe with in- gorithms needed to make good missions, a number of tech- ferently to changing fading con-
telligent center” to a more use of all the possibilities pro- niques are used to provide the ditions and packet deliver y
datacom-like “smart edge” can vided by fast scheduling will mobile terminal with a better speeds.
yield better results. be more complex than those chance of receiving the data Similar problems were seen
Overall, the data rate has in- implemented by existing RNC correctly. For users with a high in the early days of the Internet,
creased sevenfold, the response software, but those decisions coding rate, simple chase com- when interactions between the
time was reduced by 80 percent have to be made within a bining may be used, which sim- different layers of the protocol
and the algorithms, scheduling millisecond. ply repeats the packet. For us- stack led to less-efficient band-
and complexity have increased When link errors occur, data ers with a low coding rate, in- width use than expected. Tech-
niques were developed to over- this requires multiple scenarios user coding and modulation ad- efficiency of HSUPA, it is highly
come the problems and in- to be evaluated under different aptation. With a large number likely that HSDPA will be im-
serted into terminal equipment loading conditions. As a result, of processing elements, it be- paired in applications that
and infrastructural systems to architectures that maximize comes possible to dedicate pro- have more balanced bandwidth
bring performance back up to f lexibility will be key to effi- cessing and buffer resources al- needs.
their expected levels. cient HSDPA implementation. most on a per-user or per-func- HSUPA puts even more
If a scheduler is not designed tion basis. For example, one strenuous demands on the base-
to react to problems, operators Granularity needed processor may collate informa- station design and processing
may see some users with ter- Processing granularity will be tion for a processor that just electronics will have to deal with
minals that are able to handle an major consideration for the runs an advanced scheduling al- a much more complex decode
high-speed transfers starved of efficient implementation of an gorithm. This allows the proces- environment in the same way
bandwidth, while other users HSDPA-compliant base station. sor to perform scheduling deci- that HSDPA demands much
with less-capable systems use Systems based on a small num- sions all the time. This will yield more of the terminals in deco-
up too much of the HS-DSCH ber of high-performance DSPs much lower latencies than a sys- ding. HSUPA means moving
bandwidth. Such a situation tend to demand large buffers tem where scheduling is shared control functions from the RNC
will see much lower data use and work on large groups of data with other tasks on a general- to the Node B. As is the case for
than expected. A more intelli- at any one time to reduce the purpose processor or DSP. HSDPA, these will likely break
gent scheduler that watches for overhead of switching between many installed architectures.
changes to channel and termi- tasks. This makes things Future-proofing Given the speed that these
nal conditions—and schedules “clumpy” with high latency. A flexible, software-based de- changes are arriving, having a
packets for terminals that can However, such a coarse-grained sign will be vital for future im- f lexible or upgradeable plat-
receive at higher data rates— approach to task scheduling is a provements to the W-CDMA ser- form is important.
will improve overall revenue. poor fit for algorithms such as vice offering. HSDPA is an un- HSDPA significantly im-
However, the need to sup- scheduling that need low la- balanced system with a maxi- proves the quality and perfor-
port different quality-of-service tency to work effectively. mum of 14Mbps on the downlink mance of wireless data for 3G,
(QoS) contracts with each ter- Advanced silicon processes and 2Mbps on the uplink from with a corresponding impact on
minal will further complicate available today make it possible the terminal to the network. the operator’s profit. Changes
the situation for the scheduler, to implement hundreds of pro- That can be a concern, as TCP to the modulation, architecture
as it cannot simply deny band- cessors on a single chip with can easily be “uplink-choked” if and networking control algo-
width to a terminal with a high distributed memory blocks and acknowledgments are slow, re- rithm are all required. How-
QoS setting just because it hap- an interconnect structure that ducing the downlink rate. ever, despite some claims, this
pens to be in a poor reception efficiently delivers data needed Release 6 of the 3GPP speci- upgrade is not simple and many
area or is unable to react to implement the required feed- fication will change that by in- base stations will require new
quickly enough to the data it back paths. Protocols such as troducing high-speed uplink extensive hardware if they are
receives from the base station. HSDPA, as with earlier versions packet access (HSUPA). This al- to deliver on the potential. Im-
As well as allowing for evo- of W-CDMA, work well on par- lows users to take advantage of mediately following HSDPA is
lution in scheduler design, in allel-processing architectures, faster uplinks with lower latency its counterpart for the up-
many cases, it will be desirable as many different processes when sending large files or e- stream. This also has great ad-
to have different scheduling need to happen at the same mails. That in turn improves the vantages, but likely requires
policies in action at different time (Figure 2). efficiency of the link, increasing further hardware changes. Car-
times of the day or tuned for cer- Fine-grained control will be effective throughput even if the riers should plan for the major
tain locations, such as an air- necessary to implement features modulation has not changed. opportunity, but also minimize
port waiting lounge. Testing such as fast scheduling and per- Indeed, without the improved the disruption.

Вам также может понравиться