The Fairchild 670 sends audio through an input transformer, a variable-mu tube gain stage, and an output transformer before it reaches the next device. Most of its remaining circuitry belongs to the sidechain, which watches the output and generates the control voltage that changes gain.
Calling the result “tube warmth” throws several different mechanisms into one bucket. The 6386 tubes perform the gain reduction themselves, the push-pull arrangement changes which distortion products survive, and the feedback sidechain decides how the gain cell responds as level rises. The transformers also sit directly in the audio path, so bypassing compression does not mean bypassing the hardware.
This matters because the gain cell is not perfectly linear at every control voltage. A circuit model of the 670 found gain-reduction-dependent Fairchild distortion, with distortion increasing as the modeled limiter applied more gain reduction. The same work found only odd harmonics in its pure-tone test because the balanced topology cancelled the even harmonics.
Tube color in a 670 is therefore not a fixed coating added equally to quiet and loud passages. The nonlinear behavior can change with the amount of gain reduction, so a section being worked harder may acquire a different harmonic balance from a section barely touching the limiter. Two settings with similar output loudness can sound different simply because one asks more from the gain cell.
Push-pull operation also explains why reducing the whole machine to “lots of tubes” is misleading. Balanced stages can cancel distortion components that a single-ended tube stage would retain. The tube type, bias, matching, circuit symmetry, and amount of control voltage all matter before the word warmth tells you anything useful.
Driving a transformer harder can also move it away from ideal behavior, especially toward low frequencies where core flux becomes more demanding. This does not mean every bass-heavy signal automatically produces obvious saturation. It means a hardware recreation can use the right tube topology yet still behave differently if its transformers, source impedance, loading, or operating levels depart from the original design.
The control path matters just as much. The 670 is a feedback compressor, so its sidechain observes the amplifier output after gain has already been changed. It then rectifies and filters that information into the negative control voltage sent back to the 6386 stage.
Feedback makes the compressor’s action dependent on the result of its previous action, not merely on an untouched copy of the input. The sidechain itself is a substantial tube circuit with its own transformer, diode bridge, threshold network, and current limits. Replacing that section with a generic detector while keeping a tube audio path can preserve the broad idea of variable-mu compression without reproducing the same control behavior.
Timing then determines how quickly this changing gain reaches the music and how it lets go. The Fairchild 670 time constant settings control that envelope behavior, but timing alone cannot reproduce the full machine. A clean digital compressor can copy attack and release values while lacking the 6386 transfer curve, transformer loading, feedback detector, and level-dependent distortion.
This is why two Fairchild-style processors can share the same six-position timing layout and still sound noticeably different. One may substitute tubes, simplify the sidechain, use different transformers, alter operating levels, or model only selected nonlinearities. Matching the front panel is the easy part.
Listen with compression nearly disengaged first, then increase input drive while matching output loudness. Changes heard before substantial gain reduction point toward the audio path and operating level, while changes that grow with gain reduction reveal more of the variable-mu stage and control loop. The distinction is useful because the 670’s character is not one effect waiting behind one knob.
Calling the result “tube warmth” throws several different mechanisms into one bucket. The 6386 tubes perform the gain reduction themselves, the push-pull arrangement changes which distortion products survive, and the feedback sidechain decides how the gain cell responds as level rises. The transformers also sit directly in the audio path, so bypassing compression does not mean bypassing the hardware.
The 6386 gain cell changes character while it compresses
A 6386 is a remote-cutoff dual triode whose amplification factor changes as its grid voltage becomes more negative. Fairchild used four 6386 tubes per channel, with their triode sections arranged as parallel halves of a balanced push-pull gain stage. Compression therefore happens inside the amplifier carrying the music rather than in a separate attenuator placed around an otherwise fixed tube stage.This matters because the gain cell is not perfectly linear at every control voltage. A circuit model of the 670 found gain-reduction-dependent Fairchild distortion, with distortion increasing as the modeled limiter applied more gain reduction. The same work found only odd harmonics in its pure-tone test because the balanced topology cancelled the even harmonics.
Tube color in a 670 is therefore not a fixed coating added equally to quiet and loud passages. The nonlinear behavior can change with the amount of gain reduction, so a section being worked harder may acquire a different harmonic balance from a section barely touching the limiter. Two settings with similar output loudness can sound different simply because one asks more from the gain cell.
Push-pull operation also explains why reducing the whole machine to “lots of tubes” is misleading. Balanced stages can cancel distortion components that a single-ended tube stage would retain. The tube type, bias, matching, circuit symmetry, and amount of control voltage all matter before the word warmth tells you anything useful.
The transformers and feedback loop shape the operating conditions
The audio path itself is short. An input transformer feeds the 6386 stage, which drives the output transformer directly. Those transformers provide impedance conversion and isolation, while their real losses, bandwidth, phase behavior, and magnetic limits make them part of the electrical system rather than transparent connectors between tubes.Driving a transformer harder can also move it away from ideal behavior, especially toward low frequencies where core flux becomes more demanding. This does not mean every bass-heavy signal automatically produces obvious saturation. It means a hardware recreation can use the right tube topology yet still behave differently if its transformers, source impedance, loading, or operating levels depart from the original design.
The control path matters just as much. The 670 is a feedback compressor, so its sidechain observes the amplifier output after gain has already been changed. It then rectifies and filters that information into the negative control voltage sent back to the 6386 stage.
Feedback makes the compressor’s action dependent on the result of its previous action, not merely on an untouched copy of the input. The sidechain itself is a substantial tube circuit with its own transformer, diode bridge, threshold network, and current limits. Replacing that section with a generic detector while keeping a tube audio path can preserve the broad idea of variable-mu compression without reproducing the same control behavior.
Compression curve and timing finish the fingerprint
The 670 also changes ratio as signal level moves through its compression curve. It does not behave like a modern compressor set to one ratio that remains mathematically fixed once the threshold is crossed. AC and DC threshold adjustments influence threshold, knee, and slope together, so the gain cell can move from gentle compression toward much firmer limiting as level increases.Timing then determines how quickly this changing gain reaches the music and how it lets go. The Fairchild 670 time constant settings control that envelope behavior, but timing alone cannot reproduce the full machine. A clean digital compressor can copy attack and release values while lacking the 6386 transfer curve, transformer loading, feedback detector, and level-dependent distortion.
This is why two Fairchild-style processors can share the same six-position timing layout and still sound noticeably different. One may substitute tubes, simplify the sidechain, use different transformers, alter operating levels, or model only selected nonlinearities. Matching the front panel is the easy part.
Listen with compression nearly disengaged first, then increase input drive while matching output loudness. Changes heard before substantial gain reduction point toward the audio path and operating level, while changes that grow with gain reduction reveal more of the variable-mu stage and control loop. The distinction is useful because the 670’s character is not one effect waiting behind one knob.
