CNN vs. FIR Filter: Visualizing Neural Network Weights for Signal Processing

Skeptical of ‘Black Box’ AI? We prove that a 1D Convolutional Neural Network (CNN) is mathematically identical to an adaptive FIR filter. Includes Python code to train a CNN to reinvent the Matched Filter and visualize its frequency response.

Simulate matched filter system with SRRC filtering

Pulse-shaping Lab path:Nyquist zero-ISI → Raised cosine → Matched filter + SRRC → Eye diagramsTool: RRC filter & bandwidth calculator Key focus: Let’s learn how to simulate matched filter receiver with square root raised cosine (SRRC) filter, for a pulse amplitude modulation (PAM) system. Simulation Model A basic pulse amplitude modulation (PAM) system as DSP … Read more

Discrete-time communication system model

Key focus: Baseband communication system and its equivalent DSP implementation (discrete time model) with a pulse shaping & matched filter is briefly introduced. If a train of pulses representing an information sequence need to be sent across a band-limited dispersive channel, the bandwidth of the channel should be large enough to accommodate the entire spectrum … Read more

Symbol Timing Recovery for QPSK (digital modulations)

Digital Modulations Lab path:BPSK → QPSK → M-PSK simulation → QPSK symbol timing → π/2-BPSK (5G NR)Tools: BER vs Eb/N0 · EVM → SNR · OFDM CP overhead The goal of timing recovery is to estimate and correct the sampling instants and phase at the receiver, such that it allows the receiver to decode the … Read more