Method Description

In this paragraph, we descript the procedures of this algorithm in detail one by one. The procedures are described follows.

(1) Build window SNR table

Before start scheduling, the algorithm first defines a term called window. Here, we denote it as w. Besides, the w has a size called window size, which is an odd number less than or equal to m and denotes it as ws. The w is a rectangle that can once cover several RBs based on ws for a single UE at a RB. At this procedure, we first place the w on RB₁ of UE₁. Second, the

21 minimum SNR in w can be taken as the possible average SNR after allocating the RB, which we predict in prior.

22

UEi. After getting one from previous procedure (“Choose starting RB to UE” or “Upper bound estimation”), where is just allocated to , the sub-program checks all the of UE_i, which denotes as . If is larger than and not equals to , the

is revised to .

(4) Adjust window SNR table

In this sub-program, it revises the window table, which is first built at “Build window SNR table.” Here, we check the UEs in N one by one. At first, it puts w on the first left side non-allocated RB of allocated RBs of UE in N. Then, if the most left side allocated RB is not the current UE getting from N, the wSNRi,j is revised as

We then slide w to the second left side non-allocated RB, and revise the wSNR_i,j as

.

The w continually slides left and revises wSNRi,j. Until the wSNRi,j is revised as

the UE stops. However, if the UE in N equals to the UE of most left side allocated RB, it first towards right looking for the most right allocated RB of this UE, and denotes the RB as . Then, the w is similarly puts on the first left side non-allocated RB, . If , it stops, else the wSNRi,j is revised as

23

Then we slide w left, and revise the wSNRi,j as

until , it stops. After the wSNRi,j at left side non-allocated RBs of UEs in N is totally adjusted, the procedure tries to revise the wSNRi,j at right side non-allocated RBs of allocated RBs, which is similar to the description described above. At first, it puts w on the first right side non-allocated RB of allocated RBs of UE in N. If the most right side allocated RB is not the current UE getting from N, the wSNRi,j is revised as

We then slide w to the second right side non-allocated RB, and revise the wSNR_i,j as

The w continually slides right and revises wSNRi,j. Until the wSNRi,j is revised as

the UE stops. However, if the UE in N equals to the UE of most right side allocated RB, it

24

first towards left looking for the most left allocated RB of this UE, and denotes the RB as . Then, the w is similarly puts on the first right side non-allocated RB, . If

, it stops, else the wSNR_i,j is revised as

Then we slide w right, and revise the wSNRi,j as

until , it stops.

(5) Choose next RB

At this procedure, the scheduler tries to determine the next RB from M to allocate. Let the first un-allocated RB on the left side of allocated RBs be , and the first un-allocated RB on the right side of allocated RBs be . The decision value of is , where allocates to . Besides, the decision value of is , where allocates to . However, if allocates to virtual UE (the term here means not allocated to any real UE), the decision value of will be the maximal among all the s, where N and available to be scheduled at . Similarly, if allocates to virtual UE, the decision value of will be the maximal among all the s, where N and available to be scheduled at . The scheduler then picks as next RB to allocate if the decision value of is larger than or equal to the decision value of , else picks However, if there is no , it returns directly, and returns if there is no . Finally, if

25

there is no remaining RB to schedule, the algorithm terminates at here.

(6) Choose available allocated UEs available UEs, which determines at “Choose available allocated UEs”, respectively, and

26

allocates to the UE whom has maximal upper bound. In (1) and (4) of “Choose available allocated UEs”, the available UEs are . The available UEs of (2) and (3) are picked at “Choose available allocated UEs”. Here, let the UEs available to be pseudo-allocated at left side of allocated RBs be , which belongs to N and not includes the UE, whom obtains the most right side allocated RBs, if the UE is blocked at left side by another UE. Besides, let the UEs available to be pseudo-allocated at right side of have been completed, the scheduler chooses the maximal bound of this pseudo-allocating. It then picks the second highest priority RB and does the same thing, and stops until all the RBs in sorting results have been picked. After all prediction is completed, each UE available to be allocated at has upper bound estimation. The scheduler then real allocates to the UE, whom has maximal upper bound among available UEs. Finally, it removes from M to M’, and if this allocation cause the UE, whom obtains , blocked, then removes the UE from N to N’.

27