TR-606: circuit tracing, revision differences and shared metallic voices

Technical research · edition 2 · reviewed 2026-10-24

Roland TR-606 service schematic, sheet 5: six Schmitt-inverter source circuits. Image source. View full-size image

What we found

Our 606 investigation replaced a machine-wide pitch multiplier with component-specific targets. Its metallic oscillators, snare and two toms have different RC networks, and a marked manufacturing-lot resistor change materially alters one oscillator. The most useful insight is that shared topology does not justify shared coefficients or a universal “606 scale.”

We traced the service schematic and derived isolated modal and filter targets. They provide independently checkable building blocks, while leaving trigger coupling, nonlinear gates and real-unit calibration explicit.

The kick has two component-derived modes

The two bridged-T feedback networks share their trigger drive and feed the output through different resistors. The inspected scan gives:

For an ideal equal-capacitor network, our engineering derivation uses omega0 = 1 / (C sqrt(Rbridge Rground)), alpha = 1 / (Rbridge C), and fd = sqrt(omega0^2 - alpha^2) / (2 pi). The resulting damped frequencies are about 59.85 and 128.04 Hz; lower T60 is 263.05 ms, with the upper decay rate 4.5333 times faster. The impulse-coefficient and output-resistor ratio gives an upper mix of 0.57037.

Implementation uses these poles in the existing modal engine. It does not solve the complete circuit: a shaped finite trigger has a different modal excitation from the ideal impulse used for these ratios. Tune and Decay scale the reference; Body controls the upper resonance and Damping adds upper decay. Sweep, Punch, Click, Harmonics, Drive and room remain explicit synthesis extensions. The Drumatix preset disables those extensions. Acid knock remains a designed variation. The fixed sine-phase excitation and retrigger crossfade are still approximations.

Evidence: research/drums/HARDWARE_ACCURACY.md.

The oscillator bank and the revision-dependent frequency

IC16 is labelled HD14584BP, with pin 14 at +5 V and pin 7 at ground. Each oscillator is an RC Schmitt inverter, rather than a sine mode. Its output is summed through a 150 kohm resistor. Reference order below follows the schematic, not a fabricated list of 606 factory tuning values.

The R226 annotation reads 84 Lot ~ 680K. Treat that as a manufacturing-lot change. It is not evidence that every unit built in a particular calendar year has the later resistor. Both values must remain traceable during calibration.

The summing node has R210 = 1 kohm to ground. Ignoring loading by the following filters and inverter output resistance, each oscillator's contribution is (1/150k)/(1/1k + 6/150k) = 1/156 of its voltage. That gain is an engineering derivation for the isolated passive mixer, not a fitted audio normalization.

Let L = ln((Vhigh-Vlower)/(Vhigh-Vupper)) + ln((Vupper-Vlow)/(Vlower-Vlow)). Each ideal oscillator has f = 1/(R*C*L). Duty also depends on the switching thresholds. Thus the schematic's R/C values alone do not fix an absolute frequency. Nominal chip thresholds, output loading, component spread and the unit revision are consequential.

For scale only, the table below applies Vlow=0, Vhigh=5, Vlower=2.1, Vupper=2.7 V, giving duty 47.9805%. Those are published MC14584B typical values, not measured HD14584BP values from an original 606. Their use is an explicit proxy assumption; they must not be advertised as calibrated 606 frequencies. ON Semiconductor/Motorola datasheet, electrical characteristics, original table on mirror.

The relative 1/(RC) structure and revision choice are firmer evidence than an arbitrary machine-wide pitch scale. Prefer a measured chip/source sweep when available. No original 606 oscillator capture is currently in the reference bank.

Evidence: research/drums/606-CIRCUIT-NOTES.md.

Bias resistors change the high-pass result

Useful isolated small-signal targets can be derived from the following component sets. Here the DC supply is AC ground, so two bias resistors to the rails appear in parallel. Ignoring the upper bias resistor would produce the wrong Q and frequency, and can even suggest an unstable high-pass stage incorrectly.

The gain-two op-amp stages use equal 39 kohm feedback/bottom resistors (R179/R177 or R205/R206). Assuming ideal buffers, their nodal equations yield:

The isolated targets are Q32 f0=2665.946 Hz, Q=0.75378, Q40 2181.228 Hz, Q=0.75378, and IC12A/B 8495.062 Hz, Q=1.13545. These are natural frequencies, not necessarily the spectral peak of the whole voice. Source impedance, coupling capacitors, finite transistor gain, gate impedance, summing and output loads alter the full transfer. Do not simply cascade these isolated filters and claim a complete circuit simulation.

Evidence: research/drums/606-CIRCUIT-NOTES.md.

Snare and tom endpoints are conditional calculations

The resonators include transistor-dependent branches; their effective resistance changes with circuit state. The following is a component ledger and two ideal limiting calculations, not a calibrated pitch sweep or static replacement model.

For the isolated equal-capacitor bridged-T, use omega0=1/(C*sqrt(Rbridge*Rground)), alpha=1/(Rbridge*C), fd=sqrt(omega0^2-alpha^2)/(2*pi) and T60=ln(1000)/alpha.

Do not treat these endpoint calculations as measured initial/final hit pitches. Q17/Q307/Q308 conduct continuously, not as binary ideal switches. Trigger drive, capacitor history, transistor bias and the other feedback/coupling paths must be solved before assigning an envelope trajectory. The lower/higher tom distinction already depends on different capacitance and resistance; multiplying one generic tom by an equal-tempered interval loses that structure.

The tom sub-board also contains Q305/Q306 and Q309/Q310 noise/control paths, shared rail decoupling and coupling networks. Their nonlinear interactions have not been solved here. The snare's Q30 and Q32 branches likewise require routing and envelope calibration beyond isolated pole tests. A clocked PRNG can supply a provisional noise input, but it is not the physical transistor noise source.

Evidence: research/drums/606-CIRCUIT-NOTES.md.

Applying the findings

Use the component ratios to define a neutral reference, then expose creative controls as changes relative to it. For an ideal equal-capacitor bridged-T, compute omega0, damping and T60 from the actual resistor/capacitor values rather than tuning an unrelated sine bank by ear.

When reducing a biased transistor filter for small-signal analysis, treat both supply rails as AC ground. The upper and lower bias resistors then appear in parallel. Omitting one changes Q and may even make a stable physical stage appear unstable in the derived equation. Keep revision choices alongside the coefficients so another developer can reproduce the same nominal target.

Limits and open questions

The calculated off/shorted transistor endpoints are not measured initial/final pitches. Actual conduction varies continuously and depends on trigger history. RC products in the hat envelope are time constants, not complete T60 values. The original 606 has low/high toms; a middle tom and additional tuning controls are software features. Same-unit recordings are still needed to fit the complete voices.

Primary references

• Roland TR-606 service notes
• Roland TR-606 owner's manual

Research provenance

The methods and results above are drawn from these retained research records. Dates and product versions inside the article identify the scope of each experiment; publication review does not mean the hardware or numerical tests were rerun.

• research/drums/HARDWARE_ACCURACY.md — SHA-256 9d5798d139a784814d46829caefa6c7c8c23234a8ca25780c806b53826b420b1
• research/drums/606-CIRCUIT-NOTES.md — SHA-256 bd736bdc5f1886c1d0d3d26e1c96ef0fc01abacb97113e5c366c14f5a28c4d45

Related projects

KickGenerator · SnareGenerator · HiHatGenerator · TomGenerator · CymbalGenerator

All audio and hardware research · Audio Plugins