### Table 5: Attributes for the two variants of parallel FFT.

"... In PAGE 11: ... An important aspect of the algorithm is that it scales poorly on a large number of processors [3]. We worked with two variants (see Table5 ) of the parallel implementation of FFT to overcome its poor scalability. Each variant executes FFT on multiple sets of n complex values to be transformed.... ..."

### Table 1 Number of multiplications and additions for convolution.

"... In PAGE 3: ... Convolution programs for N = 20 = 4 x 5, N = 60 = 4 x 3 x 5, and N = 180 = 4 x 9 x 5 were written in the PL/I programming language and run on the IBM System/370 Model 168 in a time-sharing system. The fac- tors were put in the optimal order according to the size of T(rJ in (14) to yield the numbers of operations listed in Table1 . The timings for the calculation of the cyclic convolution on the IBM System/370 Model 168 with the RT method and the FFT method, in floating-point arithmetic, are given in Table 2 in milliseconds.... ..."

### Table 10: Performance of the distributed FFT algorithm on the Paragon.

1996

"... In PAGE 30: ... A graph of the execution time versus the number of processors used for the model problem is shown in Figure 27. Table10 shows the execution time and the performance in M op/s for this algorithm. 4.... ..."

Cited by 4

### Table 18 Cloglog and PLS cloglog models goodness of FFt

2004

"... In PAGE 24: ....4.5.3. Generalised linear model on t1. Table 17 reports the results from generalised linear regression McCullagh and Nelder (1989) of y on age and t1. Remark Table18 shows how the goodness of FFt criteria for the cloglog and the PLS cloglog models are really similar. They lead to accept these models as the values of deviance... ..."

### Table 3: Two-level FFT on multiple processors with top-level ST Factorization.

"... In PAGE 17: ... For N = 29 29, the top level of the two-level factorization with the ST form is F29 29 = (F29 j[I23]j j lt; I26 gt; j)L29 29 29 T 29 29 29 (F29 j[I23]j j lt; I26 gt; j). The performance for perfect powers is shown in Table3 . However the code is generated, the two level form will have a loop structure in which the outer loop is sequential and the inner loop parallel.... ..."

### Table 2. Number of nontrivial multiplications and additions to compute an N-point FFT

2005

### Table I Total Numbers of Complex Multiplications for the Different FFT Algorithms with Length N

### Table 1. Representative Model Parameters for Garp Recon gurable Architec- tureThe input sequence of tasks to be executed is is the FFT butter y operation. The butter y operation consists of one complex multiply, one complex addition and one complex subtraction. First, the loop statements were decomposed into functions which can be executed on the CLU, given the list of functions in Ta- ble 1. One complex multiplication consists of four multiplications, one addition

1998

"... In PAGE 8: ... For the given architecture we rst determine the model parameters. We cal- culated the model parameters from published values and have tabulated them in Table1 below. The set of functions(F) and the con gurations(C) are out- lined in Table 1 below.... In PAGE 8: ... We cal- culated the model parameters from published values and have tabulated them in Table 1 below. The set of functions(F) and the con gurations(C) are out- lined in Table1 below. The values of n and m are 4 and 5 respectively.... ..."

Cited by 14

### Table 1: The 60 shortest propagation sequences (out of 1358) in- cluding four specular reflections and two diffractions off the edge of the panel in the Bell Labs Box with diffracting baffle. For each path we assign a unique identifier and give the corresponding path length, propagation delay and sequence of events along the path.

2002

"... In PAGE 6: ... 1 2 3 4 5 8 15 12 18 19 23 22 21 17 20 11 16 10 14 13 6 7 6 7 2 3 8 10 4 5 11 1 9 Figure 12: In order to track the effect of different propagation se- quences on the resulting impulse response, we assign a unique iden- tifier to every edge (cyan) and face (yellow) in our model. Table1 shows detailed simulation results for the test configura- tion with the diffracting baffle. Specifically, the columns list the path id, length, time delay, and sequence of scattering events, for the 60 shortest propagation paths (out of 1358 simulated).... In PAGE 6: ... The ability to relate the propaga- tion delay to the sequence of events along each path is a powerful tool for analyzing the impulse response since it makes it possible to derive correspondences between features in measured and simu- lated responses. For instance, we labeled some key features of the simulated response in Figure 10 with the corresponding path iden- tifiers in Table1 . The directly diffracted contribution (the shortest possible path) is clearly identified (#1), followed by contributions of the back and side wall occluded by the panel (#2 and #3).... ..."

Cited by 2