The Receiver Wasn’t Cleaning the Signal. It Was Making a Decision.

My initial question was: “How can the receiver restore a perfectly clean waveform?” That question made every noisy sample look like damage that had to be reversed. A better question was: “Which possible symbol most likely produced what the receiver observed?”

The receiver knows the set of expected reference waveforms defined by the modulation. Correlation measures how well the received interval agrees with a reference. The resulting value is useful as a decision statistic even if the raw received waveform still looks noisy.

This made matched filtering feel less like visual cleanup and more like organized evidence gathering. The receiver compresses many samples across an interval into information suitable for a symbol decision. A beautiful waveform is not the goal of that step; a reliable distinction between candidates is.

Our simulation compared ASK, FSK, and PSK, and each scheme required the receiver to interpret evidence differently. With ASK, the useful statistic can be compared with a threshold because the possible symbols differ in amplitude. The receiver looks for enough carrier-aligned energy to support one amplitude state rather than the other.

With binary FSK, the candidates occupy different frequencies. Comparing correlation with two reference waveforms asks which candidate frequency is better represented in the received interval. The result is relative evidence: one reference should agree more strongly than the other.

With coherent PSK, the sign or alignment of correlation can distinguish opposite phase states. The receiver is searching for the property the transmitter intentionally changed. Amplitude, frequency, and phase are not only labels for the transmitted waveform; they determine the form of the receiver's decision.

One source of my confusion was the clean waveform shown after detection. It was easy to read that image as proof that the receiver had repaired every noisy sample. In our project, the clean result represented an ideal waveform regenerated from hard symbol decisions.

Once the receiver selected a zero or one for an interval, the simulation could draw the ideal waveform associated with that symbol. That reconstruction was useful for comparing detected symbols with what had been transmitted, but it was not a recovered copy of the continuous noisy trace.

Visual cleanliness therefore said little by itself about detection confidence. Two decisions can produce equally clean regenerated symbols even if one statistic was far from its boundary and the other was close. The hard output hides some of the uncertainty that existed before the decision.

Our documented simulation used a short bit sequence and limited channel conditions. It was valuable for tracing modulation, noise, correlation, decisions, and regeneration, but it was not a general statistical proof of receiver performance.

It also was not physical hardware validation. A real communication receiver would introduce further concerns such as synchronization, channel behavior, implementation limits, and imperfect carrier knowledge. Preserving those limits keeps the conceptual lesson credible.

What changed for me was the question I brought to the receiver. I no longer expected it to undo noise sample by sample. I looked for the statistic it formed, the reference it used, the boundary it applied, and the uncertainty contained in the choice.