A High Level Performance Estimation: Modeling the Effects of Parameters on Performance Properties for a Tool Assisted SDR Development

In an library based waveform (WF) development for software defined radios, high-level performance estimation models of implementations are beneficial for converging quickly to an optimal mapping of a WF-description to a hardware platform. A model that estimates the changes in performance, e.g. bit error rate (BER), due to the configuration-parameters, e.g. input datawidth, in implementations is helpful for evaluating a mapping decision with tool assistance. Using FFT as an example, this paper introduces a general method to estimate the BER variations due to the finite word length (FWL) effects in implementations by modeling the error as a noise. Quantitatively, the root mean square error due to the FWL effects has been linked to the noise power spectral density (PSD). This leads to the modification of theoretical expressions based on the PSD of the channel noise to estimate the BER. The accuracy of the proposed model is extensively verified using 10 commercial implementations of FFT in 15 different configurations. Additionally, architectures of various FFT implementations have been analyzed and a high-level estimate of latency is provided. Simulation results demonstrate the high accuracy of the proposed models.