A homemade audio distortion analyser (10Hz - 100KHz). The tracking notch filter is based on the Tektronix AA5001 and everything else is my own design. I built it back in 2003 with a few subsequent updates/changes - the last was a firmware update to the controller Nov 2022.
It uses average detection, RMS calibrated so it hasn't kept up with the latest standards. It has a residual of about 0.0014% with a 0.5V RMS 1KHz test signal from a clone of the Tektronix SG505.
It uses NE5534 opamps in the main signal path and the tracking notch filter. The tracking notch filter is a very special design by Bruce Hofer who went on to form Audio Precision (with other ex Tektronix employees). It uses Vactrols (VTL5C4) for the notch filter frequency control with very low distortion. VTL5C4s are still available but for the last 30 years, they've been made to a different cadmium sulfide chemistry and have much higher distortion. So, this unit can't easily be replicated i.e. the VTL5C4 you buy now won't be suitable (I bought mine back in 1990s). This might have been one of the reasons Bruce Hofer went to a state variable filter with "MDAC"s (major frequency control) and FET for fine frequency control in the Audio Precision equipment.
When I was designing this unit, I've found that the rotary mechanical switches' non-linear contact resistance can introduce measurable distortion when passing the signal. So I used a Motorola MC68705P3S single chip microprocessor to scan the selection switches (with DC/pullup resistors) & then to control the various relays that actually pass the signals. These relays have high quality contacts which are far better than the readily available rotary switches.
Being a "one" off, the unit was built using veroboard. With careful layout & using ground protection tracks/planes, it achieved a noise floor of -114dBm (shorted input and with 80KHz B/W filter). Please note, this unit's input is unbalanced and thermal noise is lower compared to an electronically balanced input stage (with similar devices/topology). I should explain that whilst a balanced input will have better common mode rejection (for hum/noise), it does have one downside i.e. the thermal noise contribution of the two inputs. A designer has to work much harder for lower thermal noise in a balanced input topology...
It also includes a monitor speaker for identifying obvious problems but for better understanding of the type of distortion, you can't beat looking at the function O/P on an oscilloscope.
Anyhow, here are the pictures: