Part 6 · Data Conversion
Key point: a memoryless nonlinearity does not add noise — it adds new tones at harmonics of the input. SFDR is fundamental-to-worst-spur, and it can dominate the quantization floor.
Transfer + time · residual = the new harmonic · spectrum like a analyzer (dBc, 0 at top)
Transfer \(v(x)=a_1 x+a_2 x^2+a_3 x^3\)
Time · two cycles of \(f_0\)
Residual = output − fundamental · this is the new tone(s)
Spectrum · dBc · 0 at top · needles at \(k f_0\)
Odd \(a_3\) → only HD3 (and a little extra fundamental). Even \(a_2\) → DC + HD2 from \(x^2=(1-\cos 2\theta)/2\). SFDR is fundamental to the worst spur, not always HD3.
Try it: “HD3 only” — residual is a clean \(3f_0\) sine (three wiggles per cycle) and only the rose needle rises. “HD2 only” — residual is \(2f_0\), amber needle. “Messy” — SFDR is whoever is taller. Nothing like Part 3’s noise floor.