Turning an old PC into a MIDI-CV converter

[gstopp at fibermux.com]

     Hi list, here's another DIY project from hell...
     
     A while back I posted the idea of building my own MIDI-to-CV converter 
     based on my long-dormant PC-XT and an Analog Devices D/A chip. Well I 
     got the heart of it running and so I thought I'd post the details. Be 
     aware that I haven't yet made any music with it, but it looks like it's 
     gonna work pretty good and I figgered I'd let others in on it so they 
     could start on their own if they wanted to (or suggest better features 
     before I finish it).
     
     I had thought about a hardware-based MIDI-CV solution, then realized 
     that using a processor would be a whole heck of a lot better. But an 
     embedded solution (uP with EPROM) needs an EPROM development system, 
     and even then once it's built it wouldn't be easy to change its 
     functionality on the fly. A multi-channel MIDI-CV converter should be 
     able to control pitch and gate on several different synthesizers, or 
     control multiple CV's on one or two modular machines, or anything in 
     between. I wanted a more open-ended system. So my thoughts turned to 
     my old dusty PC-XT, with its MPU-401 still installed. I figgered that 
     all I'd have to do is put a multi-channel D/A converter on it, and 
     write some code, and I'd have a cool and cheap MIDI-to-analog 
     controller with a powerful user interface (CRT, mouse, keyboard), that 
     could do whatever I wanted it to.
     
     (The XT used to be the main sequencer in my setup, running Cakewalk 
     and Texture under DOS. Even though it would make a fine slave 
     sequencer, I have 112 MIDI channels already on the new PC. My wife 
     wants to throw the XT away to free up the garage shelf space. It's 
     really not worth much right now.)
     
     The design uses an Analog Devices AD7228 Octal 8-bit DAC which is 
     installed in an external rack-mount panel that attaches to the PC via 
     the parallel port. I chose the printer port over the serial port for a 
     couple reasons:
     
     1. The external hardware would be less complicated 
     2. The throughput would be higher
     
     The PC's parallel port provides eight data bits, plus four more 
     outputs that are used as control pins for the printer. All these 
     signals are controlled by two I/O port address locations - one for the 
     eight data bits, and another for the control lines (with four unused 
     bits). All of these signals behave simply as latched bits, so even 
     though the documentation labels the control pins as such things as 
     "data strobe" or "select", I didn't follow these functional 
     assignments since I'm not going to use standard BIOS calls or printer 
     drivers.
     
     Since there are four control lines, the DAC is eight channel. Three 
     control lines function to select one of eight channels, and the fourth 
     control line is the data strobe to write in the data. A 74HCT259 octal 
     addressable latch is used to provide eight latched bits for eight 
     V-gate outputs. Since MIDI note values range from 0 to 127 (a range of 
     slightly over ten octaves), only 7 of the 8 data bits into the DAC are 
     used, so I simply grounded the D0 input on the DAC and ran the lowest 
     7 bits from the parallel port into the highest 7 bits on the DAC. The 
     MSB of the data goes into the addressable latch. Therefore writing a 
     value between 0 and 127 to a DAC channel will create an analog voltage 
     with its corresponding gate off, and writing a value between 128 and 
     255 to the channel will create an analog voltage with its gate on. 
     This corresponds to a MIDI "note-off" and MIDI "note-on".
     
     The DAC requires a low-impedeance level on the Vref input. For this I 
     used a 78L05 regulator with a trimpot to set it to the required 
     reference level for 1V/octave semitones per bit (slightly over 10 
     volts). If you don't do this the reference voltage will be loaded down 
     depending on the state of all the DACs and the voltages will have 
     errors.
     
     Each CV output from the DAC goes into a lag circuit, into a 
     level-shifting buffer, into a scale control. Each gate output from the 
     addressable latch goes into a comparator to bring the level up to +12 
     volts from +5 volts, to provide the gate output and to drive the gate 
     LED. Therefore each CV and each gate output require one op-amp. I'm 
     using four LM324's total. Each channel will have the following:
     
     Gate LED
     Lag Control
     Range Control (DC offset trimpot)
     Scale Control (attenuation trimpot)
     CV output jack (1/4")
     Gate output jack (1/4")
     
     Power supply requirements are as follows:
     
     +15    - DAC, Vref, CV op-amps
     +12    - Gate output comparators
     +5     - 74HCT259 latch
     GND
     -15    - CV op-amps
     
     I chose to put the gate output comparators on their own regulator to 
     avoid glitching the supply that the CV's run on.
     
     To write data into the DAC, I use the following algorithm:
     
     PortReg_1 = Data register
     PortReg_2 = Control line register
     Address   = DAC channel, 0 to 7
     Value     = DAC data, 0 to 127
     
     Send (Address, Data)
     
     OUT PortReg_2, Address plus data strobe high 
     OUT PortReg_1, Data
     OUT PortReg_2, Address plus data strobe low 
     OUT PortReg_2, Address plus data strobe high
     
     RETURN
     
     The first OUT sets up the address lines so they'll be stable for the 
     third OUT.
     
     The second OUT sets up the data lines.
     
     The third out strobes the data. Data strobe low writes the data into 
     the DAC (the DAC is transparent when strobe = low, and latches data on 
     the rising edge of the strobe).
     
     The fourth OUT brings data strobe high again to latch the DAC, as well 
     as leave things in a quiescent state for the next event.
     
     To test the DACs I used a test program in QuickBasic, with nested 
     FOR/NEXT loops to write ascending staircases to all DACs 
     simultaneously. This resulted in eight rising 10 volt ramps, with an 
     overall repeat rate of 283 milliseconds. This multiplies out to 3,618 
     note events per second (8 x 127 x .283).
     
     Also I measured the note duration writing to a single DAC channel. 
     This measured 246 microseconds per note, which comes out to 4,065 
     note events per second.
     
     All of these timing measurements will be machine dependent, I think. 
     This was done under QuickBasic under Windows on what appears to be 
     some run-of-the-mill 386/25 here in the lab. It'll probably be slower 
     on my XT, especially as I add more smarts to the code, but it's 
     plenty fast enough already so I'm not too worried.
     
     I don't have any plans to add linear VCO control capability, as all 
     my analog synthesizers are exponential. Somebody else can tackle that 
     part.
     
     Anyway in another post I will bring up the subject of how to parse 
     MIDI data to control the DAC under any conditions, plus muse on all 
     the possible CV scenarios and their corresponding programs. Lemme 
     know what you think guys.
     
     - Gene
     gstopp at fibermux.com