Sunteți pe pagina 1din 5

Resource Allocation for the LTE

Physical Downlink Control Channel


Patrick Hosein
Huawei Technologies Co., Ltd.
10180 Telesis Court, Suite 365, San Diego, CA 92121, USA
Tel: 858.882.0332, Fax: 858.882.0350, Email: phosein@huawei.com

AbstractThe next generation of wireless networks will simultaneously support many data and voice users. These users or
UEs (User Equipment) will share the uplink and downlink radio
resources and so, in each frame, the basestation must signal
which UEs are being allocated resources and the location within
the frame of the associated data. In the case of the Long Term
Evolution (LTE) wireless standard, this signalling is performed
over the Physical Downlink Control Channel (PDCCH). The
amount of resources allocated to the PDCCH can be varied.
However, if the allocated amount is too small then the uplink
and downlink data schedulers will not be able to schedule all
UEs that need to be served but if the allocated amount is too
large then resources that could have been used for transmitting
data are wasted. Two issues must therefore be addressed, the
determination of the number of resources to be allocated to the
PDCCH and, given this allocation, how should UEs be efficiently
signalled over the PDCCH. Our focus in this paper is on the
second issue but we also briefly address how one can approach
the first issue. We propose a simple but near optimal algorithm
for making PDCCH allocations within the subset of resources
reserved for its use. We then illustrate, through simulations, the
effectiveness of the approach.

I. I NTRODUCTION
The Fourth generation of wireless networks such as the
Long Term Evolution (LTE) standard being developed by the
3GPP standards body as well as the 802.16 suite of standards
being developed by the IEEE standards body, use OFDMA
(Orthogonal Frequency Division Multiple Access) radio technology in the downlink. In the case of LTE [1], [2], [3],
the Physical Downlink Control Channel (PDCCH) is used in
each frame to signal which UEs have been allocated resources
for uplink and downlink transmissions. One to three symbols
are allocated for all control channels. The PDCCH uses all
resources that are not assigned to the other control channels.
The PDCCH structure requires that each UE performs blind
decoding of a set of candidate channels to determine which,
if any, contains its signalling information.
The resources not used by the control channels are available
for transmission of data in the downlink direction. The data
transmission is performed over the Physical Downlink Shared
Channel (PDSCH). More resources allocated to the PDCCH
means more UEs can be scheduled but less resources are made
available for the transmission of the actual data. Two problems
must therefore be solved, how many resources (i.e., 1 to 3
symbols) should be allocated to the PDCCH and how should
UEs be allocated to these resources (i.e. PDCCH scheduling).

This can be solved as a joint optimization problem but, because


the number of options for the resource size is small (3 options),
then a simpler approach is to evaluate the performance for
each of the three cases and pick the one that provides the best
performance. Hence in this paper we focus on the PDCCH
scheduling problem for a given allocation of control channel
resources.
In order to compare PDCCH schedulers we need to determine suitable performance metrics. A perfect PDCCH
scheduler would be one that is able to continuously allocate
resources to each UE that is chosen by the downlink and
uplink schedulers up to the point where all resources for
the PDCCH, PDSCH and PUSCH (Physical Uplink Shared
Channel) simultaneously become exhausted. However, because
of the mechanism used for allocating PDCCH resources, it is
possible that we can no longer allocate PDCCH resources to
a UE even if such resources are available. In such a case
the corresponding UE is blocked while possibly still being
able to serve subsequent UEs. As more UEs are blocked the
performance of the data schedulers decreases because such
UEs cannot be served. Hence this blocking probability is of
interest together with the utilization of bandwidth and power
resources.
Note that, for each symbol that is allocated to the control
channel, the basestation can vary the power allocated to
each subcarrier (power control). Ideally it would be best to
uniformly allocate power over all subcarriers so that when all
subcarriers become exhausted the power simultaneously becomes exhausted. However, because of the constraints imposed
on the PDCCH scheduler, it may sometimes be more efficient
to reduce the number of subcarriers needed for a particular UE
by increasing the power for the reduced number of subcarriers.
In so doing it becomes easier to find suitable resources for the
particular UE. Hence we will also investigate how we can use
non-uniform power allocation to further improve performance.
In the next section we provide the model we use for the
PDCCH scheduler. In particular we provide the constraints that
are imposed due to the way that PDCCHs are located within
the available resources. We then provide various PDCCH
schedulers starting from the most straightforward approach
followed by more sophisticated approaches that involve adjustments to allocated powers. We then provide simulation results
to illustrate the performance differences of these algorithms.

978-1-4244-5626-0/09/$26.00 2009 IEEE

N = 38 CCEs
L=4
Candidate 2

Candidate 1

Potential Starting Points

Fig. 1.

Start Point

Search Space for an Illustrative Example

II. T HE PDCCH R ESOURCE A LLOCATION M ODEL


We assume that the number of PDCCH resources (Control
Channel Elements or CCEs) and total PDCCH power are
known for each subframe. Multiple CCEs (1, 2, 4 or 8) can be
aggregated to achieve different coding rates. The aggregation
level required for a particular UE depends on the channel
conditions of the UE and the size of the scheduling grant that
must be used for the UE. Each UE is allocated a candidate set
of PDCCH channels for each possible aggregate level. The
UE must then blind decode each of these possibilities for
each of the (feasible) message formats to determine if it was
allocated PDSCH or PUSCH resources. The starting point for
the candidate set for each aggregate level changes for each UE
in each subframe. However, the new starting point is always
known by both basestation and UE. For each scheduled UE
the basestation must determine a suitable aggregation level and
then must use a free channel from the corresponding candidate
set to signal the location and formatting of the transmitted data
The PDCCH scheduler must satisfy certain constraints. Let
N denote the number of CCEs available for PDCCHs in the
present frame. The set of PDCCH candidates that a particular
UE has to monitor are defined in terms of search spaces, where
a search space, SL , at aggregation level L {1, 2, 4, 8} is defined by a set of PDCCH candidates. The CCEs corresponding
to PDCCH candidate m of the search space are given by

 
N
L (Y + m) mod
+i
i = 0, . . . , L 1
L
where the integer Y changes for each subframe but is deterministically determined from the value in the previous
subframe as follows: Y = (AY ) mod D, where A = 39827,
D = 65537 and Y is initially set to a UE specific parameter.
Therefore, for sufficiently large subframe indices the starting
point of a UE at a specific aggregation level is approximately
equally likely to be any of the possible starting points. If
the specified number of candidates for aggregate level L is
ML then m = 0, . . . , ML 1. The 3GPP standard sets the
following values M1 = 6, M2 = 6, M4 = 2, M8 = 2.
In Figure 1 we provide an illustrative example for the case
N = 38, L = 4, M4 = 2. We show the potential starting
points of the search space as well as the one used.
In addition to the UE specific search spaces there is a
common search space for each aggregation level. The common
search space may overlap with the UE specific search space.
For the purposes of evaluation of PDCCH schedulers we will
ignore the common search space.
We assume that utility based schedulers are used for both
uplink and downlink (e.g., see [4] for a description of utility

based schedulers). With such a scheduler, the UE with the


largest priority is picked for scheduling. We can use the same
criterion to determine which UE should next be chosen for
PDCCH scheduling. If this UE has already been allocated
PDCCH resources then we continue to the next one. Therefore the probability that the allocation is for an uplink or
downlink transmission is proportional to the number of uplink
and downlink transmissions scheduled. Since uplink rates are
limited by the UE power then uplink connections must be
scheduled more often than downlink connections. We assume
a 3/1 uplink/downlink transmission ratio.
Next we need to determine the statistics of the aggregation
level of the chosen UE. We first consider the downlink
and consider UEs along a line connecting two basestations.
Consider a UE at the center of this line (i.e. at the cell
edge) and assume that a spectral efficiency of 1 is required to
adequately serve the UE. We model the signal losses for both
the transmitted signal and the interfering signal by the inverse
of the distance to the power of 3.5. Assuming that 8 CCEs
are used for this edge UE we can determine the necessary
transmission power. We can then compute the distance from
the basestation that would correspond to a spectral efficiency
of 1 if 4 CCEs were to be used for the transmission to the UE.
This corresponds to the threshold that determines whether 4
or 8 CCEs are to be used for the UE. We can do the same
to find the threshold that determines whether 2 or 4 CCEs are
needed and similarly for the case of 1 versus 2 CCEs. Given
these distances, and assuming that UEs are randomly dropped
throughout the cell we can then determine the probability that
a UE requires 1, 2, 4 or 8 CCEs. Since the number of resources
allocated to uplink transmissions is small because of the low
UE transmission power then PDCCH aggregate levels are also
small. We therefore assume that, if the aggregation level for
downlink transmissions of a UE is 1 then it is also 1 for uplink
transmissions but for all cases it is one level less. Using the
above model we can now determine that the probability that
a chosen UE requires an aggregation level of 1, 2, 4 or 8 is
0.35, 0.25, 0.30 and 0.10 respectively.
The performance metrics of concern for a given number of
attempted UE allocations are (1) the blocking probability (the
fraction of those that were allocated PDCCH resources), (2)
the CCE resource utilization (the fraction of CCEs that have so
far been allocated) and (3) the power utilization (the fraction
of total power that has so far been allocated).
III. PDCCH S CHEDULING A LGORITHMS
In this section we evaluate various schedulers to determine
which provide significant performance with a reasonable degree of complexity. Note that there are two general approaches
to scheduling UEs. We can sequentially allocate PDCCH
resources to UEs in the same order as they are generated by the
data schedulers or we can have the data schedulers generate a
list of UEs and then find the best way to allocate PDCCH
resources to these UEs independently of their ordering. In
the former approach we ensure that the priorities of the UEs
chosen by the data schedulers are maintained when allocated

PDCCH resources. Therefore a UE that has high scheduling


priority (e.g., because of its QoS guarantees) is given higher
PDCCH priority over a UE with lesser data scheduling priority.
In this way the performance of the data scheduler is not
adversely affected. In the latter approach the priority of a UE
as determined by the data scheduler is completely ignored.
However, in the latter approach it may be easier to efficiently
pack UEs since one can make allocations based on UE
aggregation levels. Since we believe that maintenance of the
data scheduler priorities is of greater importance we prefer the
former approach which we assume.
A. Baseline Scheduler (BL)
In the baseline scheduler a list of UEs to be scheduled
are produced by the downlink/uplink schedulers for which
the PDCCH must allocate CCE resources. If the PDCCH
scheduler cannot do so then the UE is blocked and we continue
to the next UE produced by the data schedulers. After each UE
is processed we determine the resulting blocking probability,
resource utilization and power utilization. Since we assume
uniform power allocation across all CCEs then in this case
the resource and power utilizations are identical.
In the ideal case the PDCCH scheduler will allocate resources without blocking UEs up to the point where all CCEs
(and hence all power) have been utilized. In practice, however,
this cannot always be achieved. Note that this base scheduler
will provide a lower bound on performance. An upper bound
can be determined by ignoring the constraints of the PDCCH
scheduler and assume that CCEs can be assigned as long as
space is available.
B. Resource Shuffling Scheduler (RS)
In the resource shuffling scheduler an attempt is first made
to find an allocation for the UE as in the BL scheduler.
However, if an allocation cannot be made we next try the
following. For each candidate in the candidate space of the
UE, an attempt is made to reallocate all UEs that occupy
one or more CCEs in the candidate. If all such UEs can be
reallocated then this is done and the new UE is placed in the
now empty candidate. If none of the candidates of the UEs
search space can be emptied in this manner then the UE is
considered blocked and the process repeated for the next UE.
C. Scheduling with Power Allocation (PW)
In this case we consider the benefits of trading resources
(aggregation level) with power. We again simulate allocations
as for the baseline scheduler. However, if a particular UE
cannot be allocated to one of its candidates and the UE has an
aggregation level greater than 1 then we reduce the aggregation
level of the UE by a factor of 2 and increase the power
level to maintain the same error rate for the transmission. For
simplicity we adjust the power so that the Shannon capacity
for the new aggregation level (i.e. with half the bandwidth) is
equivalent to the Shannon capacity of the original aggregation
level. Having adjusted the aggregation level of the UE we
again try to find a suitable candidate from its candidate set. If

we again fail then the UE is considered blocked and we repeat


the process with the next UE.
D. Aggregation Shaping Scheduler (SH)
Note that the probability of being able to tightly pack UEs
in the PDCCH increases if all UEs have the same aggregation level. Furthermore, note that if the power is uniformly
allocated across all CCEs and all CCEs are not utilized
because of resource allocation constraints then the total power
available to the PDCCH is also not fully utilized. In this
approach we improve UE packing by reducing aggregation
levels. For example, if the CCE utilization tends to be reach
50% before the blocking probability becomes unacceptable
then we can roughly double the power assigned to those CCEs
that are allocated to a UE. Therefore in this approach, if the
aggregation level of a UE is greater than 1 then it is reduced
by a factor of 2 and the power for the allocated CCEs are
increased accordingly. Note that the power must be increased
by more than a factor of 2. After the aggregation level of each
UE is adjusted, the baseline scheduler is used to allocate CCE
resources.
E. Shaping and Shuffling Scheduler (SS)
Note that once we shape the aggregate levels of the UEs
as in SH, then we can no longer use the power allocation
scheduler (PW) since this may cause even further reductions
in aggregation levels and result in unacceptably high power
densities for the allocated CCEs. However we can still use
the resource shuffling algorithm since this approach does not
make any power allocation adjustments. Therefore in this
scheduler (SS) we first perform aggregation level reductions
as outlined in the SH scheduler and then schedule UEs using
the RS scheduler. In Figure 2 we provide pseudo-code for this
algorithm. Note that, by turning off the shaping and shuffling
functions, we obtain the baseline scheduler.
IV. S IMULATION R ESULTS
In this section we provide simulation results for the algorithms presented in the previous section. We assume that
N = 40 CCEs are available to the PDCCH which corresponds
to using 3 symbols for control with a 10 MHz downlink bandwidth. We assume that the aggregation level of the UE chosen
by the data schedulers is randomly chosen with the following
probabilities, p1 = 0.35, p2 = 0.25, p4 = 0.30, p8 = 0.10.
The candidate set sizes are given by M1 = 6, M2 = 6, M4 =
2, M8 = 2. The start point for each search space is chosen
randomly from the set of potential search spaces for each
UE. Monte Carlo simulations are performed and the resulting
performance statistics are computed. Note that the average
number of CCEs per UE can be computed and hence we can
then obtain the average number of UEs that can be placed
within the PDCCH if all CCEs are utilized. This average turns
out to be 14.0. Hence in our plots we use a maximum of 15
UEs. We believe that an acceptable algorithm would be one
that blocks, on average, at most one UE if one tries to allocate
resources for the average number of UEs that can fit within the

Parameters:
N = 40
C = 15
Q = [0.35, 0.60, 0.90, 1]
K = floor[N, N/2, N/4, N/8]
M = [6, 6, 2, 2]
F = [2.41, 2.19, 2.09]

%
%
%
%
%
%

number of CCEs
number of UEs to be scheduled
CDF of aggregation level probabilities
number of potential start locations for aggregation level
size of search space for each aggregation level
power increase factor for lowering aggregation level

Variables:
R = zeros(N)
L = zeros(C)
S = zeros(C)
B = zeros(C)
P = ones(C)
V = {}
G = B
T = 0
E = 0

%
%
%
%
%
%
%
%
%

i-th element will be set to index of the UE allocated to it


the aggregation level (1,2,3,4) for each UE
the randomly determined search space start point for each UE
if user is allocated resources set corresponding element to 1
initial power per CCE allocation for each UE
temporary set that will include UE indices
temporary vector variable
total allocated power
total allocated CCEs

Switches:
shape
= 1
shuffle = 1

% set to 0 to turn off shaping algorithm


% set to 0 to turn off shuffling algorithm

Functions:
generate_aggregate_level(Q) %
choose_random_start_point(L) %
allocate_resources(U, R, L, S)
%
%

given distribution Q, randomly generate an aggregation level


given L, randomly generate start point and return the index
% search candidate space of UE for first free candidate;
if candidate found set corresponding elements of R to UE index
return 1 if success else return 0

for U=1:C
L(U) = generate_aggregate_level(Q)
if (shape == 1)
if (L(U) > 1)
L(U) = L(U)-1
P(U) = F(L(U))
end
end
S(U) = choose_random_start_point(L(U))
if (allocate_resources(U, R, L(U), S(U)))
E = E + L(U); T = T + L(U)*P(U); B(U) = 1
end
if ((B(U) == 0) && (shuffle == 1))
for m=0 to M(L(U))-1
V = {}
G = R
if (S+m*L(U) == K(L(U)) {S = 0} end
for k=S+m*L(U) to S+(m+1)*L(U)
if (R(k) > 0) {V = V union {R(k)} end
G(k) = U
end
suc = 1
for (j element V}
for t=1 to N
if (G(t) == j) {G(t) = 0} end
end
suc = suc*allocate_resources(j, G, L(j), S(j))
end
if (suc == 1)
R = G; E = E + L(U); T = T + L(U)*P(U); m = M(L(U))-1; B(U)=1
end
end
end
end
Fraction of blocked UEs = 1 - sum(B)/C
CCE Utilization
= E/N
Power Utilization
= T/N
Fig. 2.

Pseudo-Code for the SS Scheduler

0.25

1
BL
RS
PW
SH
SS

BL
RS
PW
SH
SS

0.9

0.2

0.8

0.15

CCE Utilization

Blocking Probability

0.7

0.1

0.6

0.5

0.4

0.3

0.05

0.2

0.1

Fig. 3.

8
UE index

10

12

14

Blocking Probability

Fig. 4.

8
UE index

10

12

14

12

14

CCE Utilization

1
BL
RS
PW
SH
SS

0.9

PDCCH. Therefore this implies a blocking probability of at


most 0.07 after attempts have been made to allocate resources
for 14 UEs.

Next we investigate the CCE utilization as a function of


processed UEs for the various algorithms (see Figure 4). Here
we find that the algorithms that perform well in terms of
blocking probability, tend to have low CCE utilizations. The
reason being that, for these schedulers, power is used to reduce
aggregate levels and so the number of CCEs needed to support
the UEs is reduced.
Finally we investigate the power utilization of the various
schedulers (see Figure 5). Here we see why the shaping
scheduler works well. It trades power for aggregation level and
is better able to place these UEs within the PDCCH because
of the reduced size. In this particular scenario the SS scheduler
runs out of power when the UE index exceeds 14.

0.7

Power Utilization

We first plot, in Figure 3 the average blocking probability


as a function of the number of UEs that have been selected to
be scheduled by the PDCCH scheduler. Note that the baseline
scheduler provides unacceptable performance. Next in performance is the resource shuffling algorithm. This algorithm suffers from the fact that neither the power nor the CCE resources
are fully utilized before the blocking probability becomes
unacceptable. Next the power allocation (PW) scheduler shows
additional improvement but still unacceptable. The aggregate
shaping algorithm (SH) shows significant performance gains.
This comes at the cost of a high power utilization but as long
as the total power is not exceeded then this is not a major
issue. Finally, by shuffling already assigned UEs in addition
to doing the aggregation shaping we find that we can get
even further improvements in the blocking probability. For this
scheduler (SS) the average blocking probability at UE index
14 is 0.05 and hence it satisfies our condition for performance
acceptability.

0.8

0.6

0.5

0.4

0.3

0.2

0.1

Fig. 5.

8
UE index

10

Power Utilization

V. S UMMARY AND C ONCLUSIONS


In this paper we considered the PDCCH scheduling problem. Several algorithms were presented and evaluated. We
conclude that a combination of power allocation and resource
shuffling provides acceptable performance.
R EFERENCES
[1] 3rd Generation Partnership Project, Evolved Universal Terrestrial Radio
Access (E-UTRA): Physical Channels and Modulation (Release 8), TS
36.211 v8.4.0, Sept. 2008.
[2] 3rd Generation Partnership Project, Evolved Universal Terrestrial Radio
Access (E-UTRA): Multiplexing and Channel Coding (Release 8), TS
36.212 v8.4.0, Sept. 2008.
[3] 3rd Generation Partnership Project, Evolved Universal Terrestrial Radio
Access (E-UTRA): Physical Layer Procedures (Release 8), TS 36.213
v8.4.0, Sept. 2008.
[4] P. Hosein, QoS Control for WCDMA High Speed Packet Data, in
Proceedings of the Fourth IEEE Conference on Mobile and Wireless
Communications Networks, Stockholm, Sweden, Sept. 2002.
[5] Qualcomm Europe, System Level Analysis of PDCCH, R1-071282, St.
Julians, Malta, March 26-30, 2007.

S-ar putea să vă placă și