Peak Heath-Robinson

Among my many happy follies, I've started setting up to make some electronic music. I've been an electric musician for most of my life, but EDM isn't something I've dabbled in a lot, despite loving the sheer power of it. In all modern genre's, the synth reigns supreme!

I'm particularly interest in trying it live and without a computer as my digital audio workstation, a technique known as "DAWless." It's more performative than using Logic Pro on my Mac, the latter a process that feels like being back at work, sometimes. Using drum machines, sequencers and synths is mostly just a matter of turning on the power and pushing a few buttons. That's the theory, anyway.

The trouble is, I'd also like to record the good performances, too, but also still have the ability to tweak the knobs on the front panels in a way that fundamentally changes the tonal qualities of the sounds. Not just volume, EQ and pan, but timbre, voicings and transposition. This requires recording the MIDI data and that requires a DAW, when the idea is to be DAWless, or an old school multitrack sequencer, and those things are as much learning as a DAW. So, I thought I'd try this idea...

A circuit diagram of a device that matches MIDI signals to an audio recorder input and, from its output, back to the MIDI chain.

MIDI runs at 31250 bits per second (or Baud) which should work ok in the audio band if I run my Zoom L-12 digital multitrack recorder at 96k sample rate. Trouble is, MIDI also runs at digital levels - 5 volts, give or take. Audio recorders prefer 0dBu level, 0.775v RMS. 5v peak-to-peak is outside the headroom of most mixing desks and could "bleed" through to outher channels in analog mixers, which even digital mixers start off as at the inputs. There needs to be some levels matching at input and output. The solution is the above circuit.

Starting on the left, above the "Tape to MIDI" label, the signal generator represents a 0dBu (0.775v) square wave coming from a recorded MIDI signal being played back. This square wave is negative 0.5v peak for the "zeros" and positive 0.5v peak for the "ones" and therefore needs amplification to a signal that is a 0 to 5v peak-to-peak square wave by the op-amp, one side of an LM358 dual op-amp IC. The op-amp runs on a single +9v supply, instead of the traditional dual supply required for op-amps and has a gain of 5x. This corrects the audio signal level, while the 5.1v zener diode ensures the voltage fed to the MIDI line, represented by the short wire between the top of the zener diode and the bottom of the first rectifier diode, never excedes 5v. This gets us a nice square pulsed output from "tape." (My L-12 is a digital multitrack deck, with the bandwidth of 32 inch per second analog tape.)

If we record a MIDI signal on a low speed analog tape or a normal 44.1kHz digital recorder, we'll lose the hard edges of the MIDI data and the synths being fed this data may crash, not respond to notes or lose sync. I recommend recording digitally at 96 or 192 kHz and, if using an open reel analog recorder, use the fastest speed your tape can run at, at least 16ips, 32ips if you have that. The faster your recorder runs, which is sample rate on digitals and tape speed on analogs, the less gain and clipping you'll need in the above mentioned tape to MIDI circuit diagram for MIDI playback. The more gain you need, the more potential there is for noise to interfere with the MIDI signal, causing potential loss of bits, or even crashing some gear.

Getting MIDI onto tape is the righthand side of the circuit, from the adjacent resistor, diode and opto-coupler, onwards. As a MIDI signal is a data signal, it's square pulses, but it can be prone to line interference like mains "hum." To prevent the earth loops that can cause this, MIDI uses a "source to sink" line, with a strong positive line current, which is blocked by an opposite, strong positive line current of equal value. Current flows through the LED of the opto-coupler when there's a 0 on the line and that current is blocked by a 1. Flowing current turns on the opto-coupler's transistor, giving a 0 reading at the opto-coupler's output. The back-current of a 1 turns off the LED, switching off the transistor, creating a 1 on the output. We buffer this data signal with the other half of the LM358 at unity gain and divide the voltage output to approximately 0dBu (0.775v RMS) and feed that signal to the recorder.

I'm starting at 5x gain for the playback signal to MIDI because there's a fine balance between dropped bits from rounded pulses and dropped extra bits from noise introduced by having too much gain. With modern recorders capable of 96kHz samplerates, square waves into the recorder should result in square waves out for digital purposes. This may not be the case if you're using 48kHz samplerates and your data pulses will have very rounded shoulders on medium speed analog recorders, simply because of the bandwidth limitations of magnetic tape. On either platform, speed equates to bandwidth and bandwidth equates accuracy. That said, even 31250kHz should be reproduceable well enough on analog equipment to be able to condition the signal with careful playback gain and zener clipping of the signal.

If I have time, I'll try to breadboard this circuit soon. I have the necessary components on hand, just not sure of my time constraints. I'll report back when I can.