FFT Counts: Savings Under a Stated Model

Follow the column that is actually being compared

The handout compares CI generating/spreading with an FFT-based design. Its multiplication column uses C(N)=(N−1)² and F(N)=N log2(N)/2. At N=64 it lists 3,969 versus 192 multiplications.

These are the table’s counting conventions. They are not CPU cycles, elapsed seconds, battery lifetime or a complete communications standard. The other table columns address other resources; combining them without a common unit would create a meaningless total.

Sources for this section: [1]

An original power-of-two input

Let N=128 and apply the stated count model. Calculate C, F and the fractional saving (C−F)/C. Round the percentage only after calculating the ratio. Then state which new evidence would be needed to replace the phrase fewer counted multiplications with faster execution.

Use log2(128)=7. This exercise deliberately grants the model and its counting convention; it does not derive the cost of every possible FFT implementation.

Sources for this section: [1]

Reference answer

C=127²=16,129. F=128×7/2=448. The saving is 15,681/16,129, approximately 97.22%. The denominator is the conventional count, not the new count. Using the new count would answer a different ratio question.

For comparison, the table’s N=64 values give about 95.16%, which rounds to 95.2% as displayed. The original 128-input result is calculated from the formulas; it is not another measured row recovered from the handout.

Execution time would additionally depend on the implementation, hardware, data movement and what operations the count includes. A lower count under this model does not by itself establish lower reception error, higher throughput or less energy use in a deployed network.

Sources for this section: [1]

Keep credit and scope attached to the record

The source identifies a particular research comparison and a joint conference reference. It is not a record establishing one person as inventor of all technology inside a cellular generation. The biography discusses that separate attribution problem with a dated interview.

No standard, full conference paper or radio experiment is claimed as read or reproduced. The handout’s multiplication column is enough to support this specific learning task; it is not permission to invent claims about every other performance metric.

Sources for this section: [1] · [3]