### Table 3. Types of DAGs found in typical digital signal processing algorithms

1998

"... In PAGE 21: ... We classify a DAG as a full DAG if it is neither a tree nor a leaf DAG. As one can see from Table3 , the classi cation revealed that of all basic blocks analyzed 56% were trees,... In PAGE 22: ... Experiments with DAGs { Leaf DAG (L); Full DAG (F) 38% leaf DAGs and 6% full DAGs. From the set of benchmarks in Table3 we have noticed that the majority of the basic blocks found in these programs are trees and leaf DAGs. Another experiment was performed, this time using the DSPstone application benchmark adpcm.... ..."

Cited by 5

### Table III. Types of DAGs in Typical Digital Signal Processing Algorithms

1998

Cited by 5

### Table 3. Types of DAGs found in typical digital signal processing algorithms

"... In PAGE 20: ... We classify a DAG as a full DAG if it is neither a tree nor a leaf DAG. As one can see from Table3 , the classi cation revealed that of all basic blocks analyzed 56% were trees,... In PAGE 21: ... Experiments with DAGs { Leaf DAG (L); Full DAG (F) 38% leaf DAGs and 6% full DAGs. From the set of benchmarks in Table3 we have noticed that the majority of the basic blocks found in these programs are trees and leaf DAGs. Another experiment was performed, this time using the DSPstone application benchmark adpcm.... ..."

Cited by 5

### Table 1: Cycle Times To verify our theory that general purpose processors have good perfor- mance for signal processing algorithms, we implemented a number of pop- ular algorithms on general purpose processors and compared them to DSP performance reported in the literature. It is important to remember that our implementations may not be the best possible for the general purpose processor, but instead represent lower bounds on achievable performance. Fast Fourier Transform Our rst benchmark was a 1024 point complex FFT. Table 2 compares the execution time on the Digital 21064 Alpha AXP with the best reported times for other DSP chips [MS90].

in in the Subject line: On Digital's EASYnet: CRL::TECHREPORTS On the Internet: techreports@crl.dec.com

1992

"... In PAGE 9: ... However, most general purpose processor implementations have very short cycle times that put them ahead of DSPs in throughput. Table1 compares the instruction cycle times for several general... ..."

Cited by 9

### Table 2 : Complexity of di erent implementations of circuits for division. Latency is measured in number of cycles needed. Throughput is number of cycles pr. computation. Throughput AT is calculated both for the size of registers equal to a full adder (static register), and half the area of a full adder (dynamic register). Parallel/serial SRT optimal for division as an isolated operation The traditional parallel/serial SRT shows the best AT for larger N - independent of the size of registers. However, this is for division as an isolated operation. Analyzing division as an isolated operation only is not especially interesting, as in general digital signal processing algorithms are dominated by multiplications and additions.

1994

"... In PAGE 20: ... This conversion is responsible for the major part of the latency for these implementations. The implementations in Table2 are: 1. Non-restoring division based on a binary ripple adder, parallel/serial architecture.... ..."

Cited by 3

### Table 9: Table of best recognizer performance for the CMS and RASTA signal processors.

"... In PAGE 32: ....1.1 RASTA vs. CMS The CMS and RASTA results obtained abovein Table9 show that the traditional CMS outper- formed the RASTA signal processing algorithm by a small margin on this task. Both results are relatively good because the recognition strategy was primitive and the task was di#0Ecult.... ..."

### Table 2. NLMS algorithm is a special implementation of the LMS algorithm, which has more stable and fast converging properties. It takes into account the variation in the signal level of the input signals in selection of the normalized step size, . The convergence of the NLMS algorithm is guaranteed for a stationary process when 0 lt; lt; 2.

2005

"... In PAGE 21: ... Therefore, when the magnitude of x[n k] is large, the LMS algorithm su ers a gradient noise amplification [52, 53] problem. To overcome this problem, the normalized LMS (NLMS), shown in Table2 , was introduced. The NLMS al- gorithm is a special implementation of the LMS algorithm, which has a more stable and fast converging properties.... ..."

### Table 2: Comparisons of word correctness (%) of the three noise reduction algorithms (SS, LSA, and OM-LSA) and the un- processed signal based on the Aurora 2 database.

### Table 7: The mean and standard deviation of the error for the naive and minimal variance edge detection sets in the test set.

"... In PAGE 4: ... The results are determined for the naive portfolio of signal processing algorithms and for the minimal variance portfolio. The results are shown in Table7 . The mean er- ror for the minimal variance detector set is 8% smaller than for the naive portfolio.... ..."

### Table 2 and Table 3 give the performances of baseline (i.e., the signal was not pre-processed by any noise reduction algorithms) and that of proposed algorithm, respectively. Clean 20dB 15dB 10dB 5dB 0dB -5dB

1999

"... In PAGE 3: ...72 Average from 20 dB to 0 dB: 68.82 Table2 Baseline performance on AURORA database in word accuracy percentage. Training is on the clean portion of the database.... ..."

Cited by 2