The Ring Head and the Recording Gap

A head winding delivers microvolts, so the first stage sets the noise floor and nothing after it can recover what that stage throws away. Bring the two leads to the amplifier as a twisted pair and amplify only their DIFFERENCE: U1A, an OPA1612 wired as a difference amplifier with a gain of 100. Hum from the motor arrives on both leads alike and cancels. In ngspice, 100 mV of 50 Hz hum on both leads left 1 mV at the output, with one of the four resistors deliberately 1 % off. A single-ended stage would have amplified the same hum 100 times and clipped. C_EQ and C_EQ2, 220 pF across the two 100k resistors, are the equalisation. Read-back rises 6 dB per octave on its own, so above 7.2 kHz the amplifier falls at the same rate and flattens the top of the band. Leave both out for the sweep in the next step, so the gain is a flat 100 and you measure the head, not the equaliser. The differentiator and comparator after it recover DATA rather than audio. A recorded bit is a flux REVERSAL, which shows up as a peak in the read voltage. U2A differentiates about a 2.43 V reference, so each peak becomes a zero crossing, and U3A, an LM393, turns each crossing into one edge. R_HYS sets its hysteresis window, which has to be right both ways. Simulated with 200 µV read pulses and 20 µV rms of noise: 330k gave one edge per transition in four of five noise runs and one extra edge in the fifth; with almost no hysteresis, 91 edges for 11 transitions; at 150k to 220k the window was too wide and transitions were lost. The edges came 25 to 35 µs after each peak. Power the op-amps from two 9 V batteries as +9 V and −9 V, and the LM393 from +9 V and ground. Its output is open collector, so return R_PU to the supply of whatever logic reads the pulses. The simulation used behavioural op-amp models, not the vendors' models.