You are on page 1of 3

HSDPA Packet scheduler algorithms

The goal of the HSDPA scheduler is to maximize the spectrum efficiency of the cell while maintaining the QoS
requirements for different data services.
HSDPA scheduler resides in the NodeB. It is a fast scheduler which can make decisions on per TTI (2 ms) basis.
Although the implementation of the scheduler depends on the vendor, there are three fundamental types of
algorithms typically used: Round Robin, Maximum C/I and Proportional fair.
In general, the HSDPA scheduler typically assigns transmission turns to different users based
on any (or all) of the following criterion:

channel condition, communicated by the UE through the CQI report;

QoS parameters, such as the traffic class, traffic handling priority (THP), allocation and
retention priority (ARP), scheduling priority indicator (SPI) etc.;

UE traffic control specifics, such as the time data has been in the buffer or the amount of data
in the buffer; and

available system resources, such as transmit power, codes and Iub bandwidth.

The scheduler decides howto serve each and every one of the users of the sector by assigning
the following resources:
1. Scheduling turns: in every TTI (2 ms) the scheduler decides which user can transmit and
which cannot.
2. Number of codes: if several users are transmitting on the same TTI the scheduler will decide how many
parallel codes are assigned to each of them.
3. Power per code: if HSDPA codes are assigned to different users, the scheduler may decide to assign a
different power to each user, based on the availableHSDPApower and the specific user requirements.

Round Robin Scheduler


The Round Robin (RR) packet scheduler algorithm is the simplest one. It distributes the scheduling transmission
turns equally among all active HSDPA users, regardless of the radio channel condition and the QoS
requirements of the application running on the mobile devices. Figure shows an example of howRR works in a
system without code multiplexing. It can be observed that users in poor radio conditions (such as User 4, far
from the cell in the diagram) receive the exact same number of turns as users in perfect radio conditions (such as
User 3 close to the cell). As a result of this, the overall spectrum efficiency using the RR algorithm is not
maximized. The fairness in time sharing of the system resource creates unfairness to those UEs which are under
good radio conditions and starving for throughput.

Max C/I Scheduler


The Maximum C/I algorithm, on the other hand, aims to optimize the overall spectrum efficiency of the system
by allocating the radio resource to the UEs with the best channel conditions. Naturally, the instantaneous
channel condition of each user can vary independently, and the fast scheduling capability of HSDPA provides
the opportunity for the system to transmit on channels with the best radio condition at any time. For most types
of radio environments, such as macro cells, the correlation among different radio channels is low, and the
HSDPA scheduler can take advantage of the channel diversity, therefore increasing the overall system capacity.
A system running this algorithm has the best overall cell throughput compared to the other methods; however, it
creates unfairness to UEs under poor radio conditions, especially those at the cell edge, which may potentially
get zero or very lowthroughput. One can see that maximal overall throughput for the cell does not translate into
equal user experiences.

Proportional Fair Scheduler


The Proportional Fair Scheduler (PFS) also takes advantage of the changing radio conditions and channel
diversity, but in addition to performing an efficient allocation of system resources, it also considers user fairness.
This provides a middle ground between the RR and the Max C/I algorithms. The PFS algorithm assigns turn and
codes to the different users based on a fair throughput and time approach as defined in the following equation:
Pi=(IBi[n])/(ABi[n])
Where
Pi is the priority of user i for the current transmission turn;
IBi is the instantaneous bit rate of user i; and
ABi is the average throughput of the connection for user i
IBi can be calculated by the scheduler based on the CQI value reported by the UE, therefore

better channel conditions lead to a bigger values of IBi. However, this does not necessarily
translate into a bigger ABi value because the history of the UEs throughput is equally
important in determining the users priority in the queue.
As an example, consider that UE A is a category 6 device (capable of 16QAM) and is in a
good radio condition with a CQI value of 22. The scheduler finds at that particular location the UE can support
up to 3 Mbps. Assume that the average throughput for this UE is 2.2 Mbps, in which case the scheduling priority
value Pi is determined to be 3/2.21.36. Now consider a second user in the cell,UEB, who is a category 12
device (capable ofQPSK) under poor radio conditions with CQI10. In that condition, the scheduler can only
support up to 600 kbps for UE B; assuming that the average throughput of user B is 400 kbps, the scheduling
priority is Pi600/4001.5. In this case, the scheduler will select UE B for fairness reasons, due to its higher
priority Pi.

Figure illustrates how the PFS algorithm works with two active users. The radio conditions vary for both User 1
and User 2, as depicted by the varying data rates that each user is capable of through time. The scheduled users
in each time slot reflects the signal quality and fairness criterion noted.
The efficiency of the scheduler algorithm depends on the cell load, traffic distribution (multiuser diversity) and
the type of application being served; therefore, when comparing scheduler performance it is important to
understand the conditions under which the comparisons are made. While theoretical studies suggest significant
capacity gains for the PFS algorithm (on the order of 40100%) [7] our tests show gains of up to 40%, but these
are only possible in low mobility situations. When the mobiles move at medium speeds (50 km/h and above) the
gain is reduced to about 5%.

You might also like