– If X = Φ, then stop via simulation that our algorithm performs better than
– (i*,j*) = arg maxi∈X,j∈{1,2,...,N} bj( (Ci,j\ Hi,j)) ; MTLB when error in channel estimation is high and
– if bj Hi,jP (ci,j |Hi,j)> 0 then w∗ i*,j* = 1 channel is On/Off. Figs. 1 to 3 above show the simulated
performance in terms of average queue backlogs for Poisson
– bj* = bj* − min(Hi*, j*, ci*,j*) and X = X − {i∗ }
distribution when the channel is On/Off. The simulation
• Assign W∗ = wi*, j*
results in Fig.2 demonstrate the superior performance of
MTLB algorithm with reliability information in comparison
6. Simulation parameter with simple MTLB algorithm with channel estimation error
Before we provide the details of simulation set up let us first and no reliability information. By reliability information we
discuss the MTLB and MWM algorithm: mean probability with which estimation is correct. In MTLB
(1) MTLB Algorithm: The subcarrier assignment uses the channel is On/Off. It is clear with simulation results in
full information about the queue lengths (full QSI) Fig. 1 that we do well in terms of the average queue backlog
and binary (ON-OFF) information about the when error in channel estimation is high. From Figs. 2 and
channel: a sub-carrier is considered ON if ci, j > 3, it is clear that we gain nothing by reliability information,
cthreshold. Then MTLB policy is used for sub carrier and the performance is almost same as the MTLB even with
allocation. channel estimation error. While Figs. 4 show the simulated
= performance under the same set up with the condition that
channel is more general, i.e., Hi, j ∈ {0, 1, 2, 3}. It is clear
For state (b,ĉ). Compute an MTLB allocation W∗
from the simulations that when channel is general rather
(2) MWM Algorithm: The sub carrier assignment uses
than just On/Off we gain nothing from reliability
full queue state information (full QSI) and full
information. Our heuristic policy performs almost same as
channel state information (full CSI)
the MWM with error in channel estimation.
• X = {1, 2, ..., K}
• loop (until stop):
– If X = Φ, then stop
– (i∗ , j∗ ) = arg maxi,j∈{1,2,...,N} bjci,j ;
– if bj*ci*,j* > 0 then w∗ i*,j* = 1 else stop;
– bj* = bj* − ci*,j* and X = X − i∗
• AssignW∗ = wi*,j*
where ci,j denotes the channel capacity for user j on server i,
K denotes the number of sub carrier and N denotes the
number of user (queues).We now compare the performance
of MTLB algorithm and MWM algorithm in [5] with our
heuristic policy. We consider downlink OFDM consisting of
32 statistically independent and identical users, and 128 sub
carriers. We assume equal power distribution among all
users, and further assume that the loss of throughput due to
equal power allocation is tolerable [15]. We can represent
the number of packets per time slot the sub carrier i can Figure 1. Average queue backlog for MTLB policy with
transmit to user j, in terms of the erroneous channel gain gi,j, SNR = 0 dB
as follows:
Hi, j =