Saturday, October 24, 2009

The Hammond Novachord: Too Far Ahead of its Time

There is a great thread taking place on VSE now concerning the meticulous restoration of a Hammond Novachord. What's a Novachord? Produced by Hammond in the 1939-1941 period, the Novachord is generally considered the first practical, mass-produced synthesizer.

Novachord undergoing restoration. All photos in this post are courtesy of D. A. Wilson.

A synthesizer in 1939? Yes, although herioc measures were required, and the instrument's reputation suffered for it. Surprisingly, the 144 vacuum tubes required for the note generation circuitry were generally not the problem. The Hammond engineers realized that the instrument would experience a tube failure every few hours if they designed the tube circuits per normal design practices of the day. They severely derated the tubes in order to increase the mean-time-between-failure from hundreds of hours to thousands (a technique that would appear again in the ground-breaking ENIAC computer a few years later), cutting the heater voltage from the nominal 6.3V to 5V, and limiting plate currents to a fraction of a milliamp. Here's what 144 tubes in one place looks like:

We'll get back to the reliability issue in a bit. First, let's go over the instrument a bit. Is it really a synth? Yes indeed, and a fully polyphonic one to boot; you can play all 72 notes simultaneously. Each voice has one oscillator that produces a sort-of saw wave, an ASR envelope generator, and what amounts to a VCA. (It isn't a VCA as we know it, in terms of actually feeding it a control voltage; the VCA and the envelope generator are both part of the same circuit. Nonetheless, it is an amplifier with time-variant gain, and it responds to the envelope generation and does what a VCA conventionally does in a modern synth.) Attack is controlled by panel controls, while release is controlled by pedals.

The filter section is paraphonic. There are three resonant bandpass filters, and several other filters for general tone control, that all of the voices feed into. The filters can't respond to the keyboard; they aren't voltage-controlled (that concept would have to wait another three decades for Moog and Buchla), so cutoff is controlled manually. The tradeoff is that they are true LC resonant filter circuits, and have a very definite effect on the sound. The overall effect is somewhat like a fixed filter bank on a modular synth.

A vibrato circuit serves more or less the same purpose as chorus circuits on modern synths. The vibrato is configured so that different notes are varied at different rates, which produces more of a chorusing effect when dense chords are played. The Novachord has a built-in amp and speakers (which unfortunately are underneath the case and aimed at the floor), and connections for external Hammond tone cabinets. (Which means that presumably a Leslie could be connected, although I haven't heard of anyone doing that.)

Here's a shot of the panel, with its stylish Bakelite pointer knobs. You can see from this panel that Hammond anticipated the ways in which synth control panels would be laid out decades later. In general, the controls are grouped with the filter controls on the left, and the envelope and vibrato controls on the right. In the photo below, two of the resonant-filter controls can be seen at the left edge, and the attack time knob is just to the right of the big downward-pointing knob:

Hammond made approximately 1000 of these beasts, which was actually a considerable number for the day, especially since production only ran for about three years. However, only a handful are known to exist today, and fewer still are in operable condition. Considering that in 1939 a Novachord cost considerably more than the average automobile, one might think that the people who bought them would have taken more care with them. One thing that might contribute to the few number remaining today is that the things are massive and heavy; no doubt some were abandoned (say, when the owner of a house containing one died) because they were so difficult to move. Another factor is that production was halted rather abruptly in 1942 after the U.S. entered World War II and some of the parts became restricted to military uses only; likely some Novachords were stripped for parts during and after the war.

However, another clue might be in this photo:

The resistors connected to that elegant key mechanism are carrying a whopping 270 volts. That was the plate voltage for the voice generating tubes. Which brings us to the actual waveform generation method used for the voices. The Novachord uses a top-octave division setup, but unlike top-octave architectures today, the octave division is not done using counter circuit -- that concept was unknown at the time. Instead, what it uses is a sort of hybrid between a top-octace divider and a VCO with a sync input. As mentioned previously, each voice has an oscillator which generates a decreasing (downward-sloping) sawtooth wave. The oscillators for the top octave are free running; their frequency is set by a resistor and a capacitor having very precise values.

Oscillators for the octaves below the top octave have a mechanism that will accept a hard sync input, but only after the oscillator has passed the approximate halfway point in its cycle. The signal from the next octave above is wired to the sync input. When an oscillator begins its cycle, it jumps to its maximum voltage and begins decreasing. When it reaches about halfway, the sync input (the waveform from the next higher octave) pulses, but the oscillator doesn't accept it yet. Its output voltage keeps decreasing until the sync input pulses again, at which point the oscillator will accept a sync input. At that point, it syncs itself to the octave above. Each oscillator syncs to the oscillator an octave above it.

It should be apparent that there are a lot of timing dependencies here, not only in determining the frequencies for the top octave oscillators, but in computing when each oscillator in the lower octaves has passed its halfway point and will accept a sync input. Unfortunately, the combination of the very high voltages and the paper capacitors that Hammond used (the only good alternative in those days was mica, which was prohibitively expensive) made the instrument's oscillator circuits very sensitive to temperature and humidity changes. Obviously, drift in a top octave oscillator would cause that note to go out of tune in every octave. But what was more distressing was "octave jumping", in which an oscillator would mis-sync, and not only jump that note to a different octave, but also the same note in all lower octaves, due to the way the synchronization was chained. Some sources also suggest that the fast-cycling relays in the vibrato circuits were prone to sticking and throwing the whole instrument out of tune.

It's actually rather interesting that the Novachord was created in the first place. Laurens Hammond was a music lover, but he was also a businessman, and he had originally gone into the organ business to create an additional market for the synchronous motors he had invented. But the Novachord contains no motors; in fact, other than the keys and the vibrato relays, it contains no moving parts at all. The circuitry is completely different from the Hammond organs, and there appears to be no parts commonality. Did Hammond really dream up a full-blown concept for a synthesizer in the 1930s, decades before almost anyone else? Maybe. It's a myth that Hammond was tone deaf; in fact, he had a keen ear for music and timbre, even though he himself did not play. He had learned quite a bit about tone and harmonics from the experiments leading up to the Model A, and perhaps the Novachord was what he saw as the next step down that path. Also, possibly, the Novachord was intended to be a technological pathfinder for future fully-electronic organs.

In any event, although apparently the company discussed an improved version of the Novachord to go into production after the end of the war, that never happened. And it would be quite some time before Hammond did anything in the direction of synths again. So Laurens Hammond left the electronic music world a rather strange legacy, an intriguing but peculiar footnote in synthesizer history. One can't help but wonder how that history might have been different had Hammond pressed the Novachord idea just a bit further.

Wednesday, October 21, 2009

Finishing the Bi-n-Tic, and A Couple of Oopsies

I finished up the panel wiring last night. Here's the results:

Before I did the last of it, I reviewed the existing work and I found three mistakes. The first two were caused by my misreading of the CGS Web page on how to wire up the "excite" input, which takes divided-down waveforms from the counter and allows them to be injected back into the filter input. For some reason, I looked it and saw a four-position rotary switch, selecting one of the Q4/Q5/Q6/Q7 counter outputs, going to the switch contact on the external input jack. But in fact, it's a five-position rotary switch, and the jack is wired to the fifth position. So that will teach me to look more closely. I had to re-do the wiring of the jack, and of the first position on the rotary switch (fortunately, I had not soldered the rest of them yet). I was worried about the rotary switch since it's a plastic bodied type, but it appears to have survived. Here's how that came out:

Note all of the unused contacts. This is actually a 12-position switch, but I've got the stop set at position 5. To the right of the switch is the pot that controls the exite input level.

The other big booboo I made was that I put the banwidth pot in the panel where the resonance pot is supposed to be, and vice versa. In the picture below, the big double-gang pot (the bandwidth pot) is supposed to be at the far right:

That wouldn't be hard to fix except for one thing: remember me talking about those locator pins on the pots, a couple of episides back? Well, the big pot has its locator pin in a different place than the small Alpha pot. So in order to put these two controls where they are supposed to be, I'd have to drill two additional holes in the panel for their locator pins. At this point, I really do not want to be drilling on the panel and getting metal shavings all over the board. I can still detach the board, but the wiring makes it difficult to get it more than a few inches away. So instead, I'm going to "fix" it by putting new legending on the panel. If reality doesn't conform to one's pre-conceived notions, then change reality!

I waited until the end of the build to insert the ICs into the board. I wound up wishing I had done that before I did the panel wiring; some of the socket locations were difficult to get to, and I bent a few pins. But eventually I got them all in. I'm glad I took a photo of the bare board before I started assembly, because with the sockets in, it was impossible to see the silkscreening that shows which IC goes where. But with the photo, I was able to figure it out. The kit substituted TL071 opamps for the LF356s indicated on the silkscreen; apparently the LF356 is no longer available in a through-hole version.

The photo below shows the pads for the panel wiring to the resonance and bandwidth pots. The silkscreen didn't indicate which pads were supposed to be connected to which terminals on the pots, so I had to follow some of the traces away from the pads to see where they went, and compare to the schematic. Here's the photo:

The three pads at the right (note that we're looking at the board upside down, relative to the silkscreen) go to the resonance pot. As near as I can figure, they go to the pot's terminals in the same order as they are on the board: left (blue) goes to the left terminal, center (yellow) goes to the wiper, and right (orange) goes to the right terminal. We'll see when I try it; it may be that I've wired it so that the pot works "backwards". If so, I'll switch the left and right at the pot. To the left of these pads, there are two pairs of pads which go to the two gangs of the bandwidth pot. In each pair, the one on the right goes to the wiper terminal; the other goes to the left terminal. (I've connected the wiper and the right terminal together, as shown on the schematic, but I don't really know that that does anything.) Note the two pads to the left of the bandwidth pads, labeled "in". These are filter inputs that don't go through the input level pot. (The one that does go through the input level pot is elsewhere.) I've used one to connect the exite signal from the excite level pot, and the other is unused.

Here are a few more detail shots of the assembly. I wound up running four grounds. The filter signal input got its own ground. The other inputs share a ground, and the outputs share a ground. The jack wiring:

The grounds connected at the pads that are intended for a Eurorack 16-pin power connector. Since I'm building this for a 5U configuration, I'm using the MOTM-style power connector above and to the right, and the six ground pads on the Eurorack area made a handy place for bringing grounds to. Note there are four here; the three I mentioned above, and a fourth that goes to the excite input level pot.

The switch that I installed for the capacitor switching mux mod:

Finally, I put the knobs on (not indexed yet, just sitting there) to show what it's going to look like when it's finished:

Wednesday, October 14, 2009

Bi-N-Tic Filter: Panel Assembly and Hackery

Didn't turn on the soldering iron tonight. This was the night for mechanical work on the panel. The first thing I had to do was decide what to do about the locator pins on the pots: cut them off, or use them? Using them meant drilling additional holes in the panel. What I decided to do was cut off the pins on the pots that are soldered to the board (they aren't going to rotate anyway), but use them and drill the additional holes for the pots that are panel mounted. Here's how it came out:

Another angle:

This doesn't show from the front; the knob body covers the hole:

So I went ahead and drilled the rest of them. Holding it up to the light:

Note the butchery on the second from the left. This pot is a double pot; two on one shaft, and I neglected to note that it has a larger body and so the locator pin is further away from the center. Fortunately, the knob will still cover the mess up. The hole in the very center is the one I added for the switch that will select which outputs from the counter will be routed to the capacitor-switching muxes.

I had been a bit up in the air about what knob I was going to use on the rotary switch, so I test-fitted the switch and knobs to the panel. Here are the candidates. First we had the knob that came with the kit; it is of course popularly known as a "chicken head" knob. It's a style from the 1950s. I like retro-modern as much as the next guy, but to be honest, I just don't know if I care for the chicken head knob on this panel.

Second up, we have a knob from my knob collection. I don't know where this came from, but it is closer to the style of the other knobs. However, it's rather large, and it covers up some of the panel legending above the jacks below the switch:

Here's an interesting one. It is one from a Radio Shack two-pack that I bought in 1980 for a guitar repair job. As it turned out, I only used one, and this one has been in the original baggie since then. Here it is:

And, just for fun, the bag it came in:

Yes, it says $1.89, and that was for two. Ah, those were the days. Of course, in 1980, I was a starving college student, and 1.89 was a week's worth of Kraft Macaroni & Cheese. Anyway... what I'd really like is the style of knob that uses on its rotary switches. Here's an example:

So now it's time to assemble the pots:

A close-up the resonance and excite pots. Note that the shaft of the resonance pot is shorter:

Next step was to get the rotary switch installed. I discovered, somewhat to my surprise, that it also has a locator pin:

At first I couldn't figure out how to set the rotation stop limit -- it's a 12-position pot, and the circuit only uses 5 positions. Then I realized: the way to do it is to pry up the little washer that you see at the very base of the shaft housing. It has a little tab that sticks into a hole, and that is what sets the stop. You pry it up, rotate it to the proper position (they are marked on the body face), and then push it back down.

Having done that, I decided what the heck, and I went ahead and drilled a hole in the panel instead of cutting the pin off. I was concerned about whether or not the knob body would cover the hole, due to the radius of the switch body, but I went for it anyway. Here's what it looked like as I was trying to install the switch -- I had to drill the hole out a couple of times with progressively larger bits in order to get it to fit:

Sure enough, the body of the chicken head knob didn't cover the hole:

But that was OK since I didn't plan to use it anyway. The body of the aluminum knob covered it just fine:

Next step was to fix the paint abrasion that I caused when I drilled out the hole for the new switch. This is easily done with a bit of black modeling paint on the tip of a paper towel:

I didn't put all the knobs on yet, but I did one to see what it will look like. From the front:

And from the side:

The bottom flange of the knob is up off the paned about 3/16", which seems a bit much. But I checked with some other modules I have, and they are all in that same range. I really don't want to have to get out the hacksaw and cut 1/8" off of the shafts, although I am concerned about the resonance pot with its short shaft. It'll look funny if the resonance knob is right down against the panel and the others are all standing off of it this much.

The last task for the night was to cut the locator pins off of the pots that are soldered onto the board. This turned out to be easier than I thought; applying sideways twist with a pair of needle-nosed pliers snapped them right off. The results:

And a close-up -- look at the right edge:

Next: install the jacks and start wiring up the panel.

Tuesday, October 13, 2009

Bi-N-Tic Filter: Panel Wiring

I just finished wiring all of the panel leads to the board. Here's the results:

If you look at the right edge of each of the four pots, you can see that they have a locator pin. I'm still trying to decide if I want to use them, or cut them off. If I use them, I have to drill seven additional holes in the panel. Lot of trouble. On the other hand, if the locator pins are used, it pretty much assures that the pot bodies won't rotate behind the panel and mess up the knob indexing.

For my mod, which will provide for passing the Q4 instead of the Q3 bit from the counter to the most significant bit of the muxes, I had to solder a wire directly to that pin of the 4024 counter IC.:

It's a bit hard to tell from the photo, but that pin of the IC is bent outward and is not in contact with the socket. The yellow wire soldered to it will go to the panel switch for this mod. The output from the switch returns as shown in the next photo:

The blue wire brings the signal back to this otherwise-empty hole under the pin socket that I broke off.

One of the few things I don't like about this board is that the various points on the board, where the pads are for the wiring to/from the panel, don't have ground pads in the vicinity. This means, among other things, that you really can't use twisted-pair wiring. In fact, grounding points are kind of scarce. The only convenient place to bring out grounds is to use the set of pads that are intended to be used for a Eurorack-style power connector. It has six pads connected to ground. I've soldered in two wires, with the idea that one will be the ground for the input signal jack, and the other will be the ground for all of the control and output jacks. But I'll look at again when I put together the panel. Here's the ground wiring; note the MOTM-style power connector above:

Wednesday, October 7, 2009

Continuing on the Bi-N-Tic Filter

All of the parts are stuffed, except for the socketed ICs that haven't been inserted yet. Here's how it looks:

That blob in the upper right corner is where the VCO's expo converter is; it consists of an integrated matched pair with a tempco in contact with the top of the package. Here's a close-up:

The white stuff is heat sink compound, which should improve the thermal contact between the IC and the tempco. The board silkscreen calls for an LM394 here, but the kit came with a SSM2210, which caused me a bit of confusion until I went back and looked at the Bridechamber writeup for this module -- it specifically mentions this substitution, as having been done for improved tracking.

This gap-toothed socket is where the 4024 counter will go:

That's pin 9 that is broken off, which is the counter's bit 3 (8's) output. I am going to add a switch that will allow that line to be switched to either bit 3 or 4 (16's), which will alter the scanning pattern of the capacitor banks and should produce interesting results, I think. I'll bend that pin of the IC out so it doesn't go into the socket, and tack a lead wire directly to it to route to the switch. The return from the switch will come back to the un-contacted pad, to be routed to the cap switching muxes. Here's the underside, showing the as-yet unsoldered pin 9 pad:

Next step is the panel wiring. I'll solder in MTA-100 headers for some of it, but there are some places where there won't be enough room for the connectors.