Interview with Dave Rossum

– The founder of E-mu Systems.

Questions asked by Kenneth Abildgaard of

I read on the internet once that you got inspired to build the first Emulator after having seen the Australian Fairlight CMI at the AES show in 1980. I also want to build a lot of stuff that I see – but I always fail miserably and can’t even get my head around it.

I obviously know that you build modular synthesizers and that you created the technology for keyboards for other companies like Sequential Circuits Inc. and Oberheim Electronics.

What made you so sure you could do a sampler? Had you been experimenting with digital circuitry and RAM based technology prior to this? Did you buy a Fairlight sampler to look at when developing the Emulator – or did you do it differently – from scratch?

To understand fully, you need some background: the situation in May 1980 was that we returned from AES to find that Sequential Circuits was refusing to pay the royalties they had promised, and that we had counted on to fund the marketing of the Audity – which we introduced at the show.

We needed a product soon. Scott Wedge, Marco Alpert and Ed Rudnick had been talking on the drive back from the show, and thought that the Fairlight had one and only one good feature – sampling. We had also seen a Publison Digital Delay that had a capture mode, and the captured (sampled) sound could be played with a control voltage/gate type synthesizer keyboard.

The guys came to me with their ideas, and we had the need for a new MI product quickly to replace the lost Sequential revenue stream.

E-mu was the first company to use a microprocessor in an MI product – our 4060 polyphonic keyboard and digital sequencer, introduced in 1976. We’d done all sorts of stuff with microprocessors – the Audity had a full blown real-time operating system I’d written.

We’d built our own Z-80 development system including disk interfaces, etc. The sequencer in the 4060 used 64K bytes of dynamic RAM. And as I’ve been previously quoted as saying, “Any asshole can design digital circuits.” (Analog is a LOT harder).

We also had been consultants for Roger Linn on the circuit design for the LM-1, so we knew a bit about sampling as well. We’d played with COMDACs in the lab at E-mu as well.

The Fairlight used a separate RAM and a separate CPU for each voice. When Scott, Marco, and Ed came to me with their idea, I knew that such an approach was simply too expensive for an MI product. We’d just have another Audity-class product, competitive with the Fairlight.

So I saw that the key would be to use ONE CPU and ONE memory for all eight voices. The trick was getting the memory bandwidth to accomplish that. The solution was a combination of fast, cheap DMA chips and some FIFO buffers to give them big enough bursts so that the bus negotiation didn’t hog too much bandwidth.

So the answer is that we never gave the slightest thought anything but designing the Emulator from scratch. I was revolutionizing the state of the art – building what was in my mind, not duplicating something that I’d seen. And the hardware was the easy part.

The software was the real challenge. The Audity we demo’ed at AES had about 10,000 lines of code, which I’d written in about 3 weeks. The Emulator code base was a similar size, but rather more complex in several ways. Getting both the hardware and software into a form for demonstration at January NAMM 1981 was a real challenge. And that leads to:

How in the world did you manage to design and build the original Emulator in less than a year?

Necessity is the mother of invention. We were really motivated to turn the Emulator into a revenue stream – the survival of E-mu depended on it. Scott Wedge ran the company, including managing the lawsuit against Sequential. He could easily have assisted in the software development for the Emulator; after all it was he that wrote the sequencer for the 4060. But he knew that if he could keep me from getting distracted,

I would produce at a prodigious rate. Ed Rudnick managed the production of the modular systems, providing a trickle of cash to help pay the bills. And Marco Alpert (with the assistance of the rest of us) helped me with the functional definition of the Emulator, as well as being marketing and sales for the modulars.

As for me, I lived, ate, drank and dreamed Emulator. By September (we started the beginning of June), I had the circuit design completed, boards layed out, and some skeleton software written.

It wasn’t until late October that the bugs were out of the circuit boards. I remember when Karen (now my wife) came by returning from her work late one afternoon, and asked how my day had been. I had just that instant completed the software that allowed loops in the sounds.

So the first loop was Karen speaking into the instrument, saying “Mary had a little lamb.” And I could simply hold down the key and it would play “Mary had a little little little little lamb.” (The next loop I made, after Karen left, was me peeing in the toilet adjacent to the lab. It made it sound like I had the world’s largest bladder.)

But by the time NAMM rolled around, the software and hardware were in pretty good shape. The real challenge (as it usually is) is getting it in a box. We had hired the firm of Hovey-Kelly, who did the package for the Apple II, to design the Emulator package.

The final package was made of steel and built like a tank (and weighed about as much as one too). But in January, we just had a plastic prototype. Fitting all the circuit boards into the package, getting them reliably connected, and keeping them working (there were still a lot of kluges on the boards) was a challenge.

The funniest part of the whole thing was when Karen and I were driving down to Anaheim the night before the show with the Emulator prototype in the back of my diesel Rabbit. We were on a very remote part of the coast highway, and despite her pestering me, I had let the fuel tank get pretty low. There was fog all around, no lights anywhere, and the gauge was well below empty.

Just as the engine started sputtering, there was an exit ramp. I rolled down it, through a traffic light, and into a gas station. And …. they had Diesel fuel. We rolled to the pump, filled up, and were on our way. She was so mad that I hadn’t learned my lesson about cutting things tight (I never have).

So the simple answer is – teamwork and hard work.

I know that Stevie Wonder was one of the first pros to actually play the Emulator at the NAMM 1981 show –sampling himself I’ve heard. Apart from this small miracle for your PR department and sales I guess – why do you think sampling got so big shortly hereafter?

Well, it actually took longer than we thought it would. I remember in December 1980, we had the prototype working pretty completely. It was mounted on a big piece of plywood. And one weekend Ed and Marco had several musicians come over and sample their instruments.

When I met with them on Monday, they said if we could fix the one problem (a noisy A/D converter, which I nailed after about 2 weeks of work), we had an instrument that would revolutionize how music was made. By the time Stevie stopped by our booth at NAMM, we all were sure that was true. By the way, the guy who REALLY supported the vision was Daryl Dragon. He was the first and biggest fan of the Emulator.

The E1 listed for $10K, and our business plan said we would ship 5 a month starting in July. We did exactly that through November, but we couldn’t sell a one of December’s allocation. We’d gone through all the savings we’d established from the Sequential royalties (before they stopped paying), and were looking having to close our doors.

Music Technology, Inc of New Jersey, offered to buy the Emulator any pay royalties. But when we visted them in late December, it was clear they knew less about how to market an Emulator than we did. And they didn’t really have any deal to put on the table.

We scraped together enough to get to January NAMM. We did three things to help sales. First, I had coded a sequencer for the Emulator. We’d planned to offer it as an option, but we decided to included it on every unit, and retrofit the units already sold for free.

Next, we dropped the price from $10K to $8K. But what we later all decided was the thing that turned the Emulator around was we increased the sound library from 10 disks to 25, including some pretty hot sounds. Also, by this time we had multi-sampling working (another huge improvement over the Fairlight), and that made for some pretty good disks. At NAMM, we sold 75 Emulators, and production stayed at 25 units per month until we discontinued the E1 in anticipation of the E2.

The reason sampling works so well is that the sound is so accurate. You don’t ask (like you did in the analog or FM synth days, or even with physical modeling synths), “Is that a violin?” You ask, “Is that a Stradivarius?”

While sampling continues to struggle with being sufficiently expressive (the fact that MIDI is such a crappy interface has helped us in this sense, because nobody gets to be anywhere near as expressive as they should be thru MIDI), that accuracy is so impressive that it’s hard to give it up once you’ve heard it.

The other thing sampling had on its side, and this was something I knew as we developed the technology, is Moore’s law. The quality of the samples (at least for the first couple of decades) depended on the size of the memory, and memory kept getting cheaper and cheaper. So without a lot of fundamental research, sampling kept getting better and better.

The Emulator II series of samplers is probably the most popular pro sampler of the eighties. And the mighty Emulator III followed shortly after. The Emulator 4 series is nothing but brilliant and had that ‘4x wow’ factor when they were first released; I mean a maximum of 128mb RAM and a maximum of 128 voices was unheard off. Not to mention the really cool and varied selection of digital filters and marvelous modulation possibilities.

Tell me honestly; did you really know how far from the competitors you guys really were – and were you deliberately going for this ‘Way Ahead of the Competition’ strategy?

Yep, we knew, it was on purpose. But it frustrated the heck out of us that very few other people noticed.

The “H-chip” which did the filters was a huge step forward. But it was really hard to get people to be sufficiently creative with them. And the “G-chip 2.0” that powered the E4 (and Creative’s SB Live and X-Fi cards too) has yet to be equaled, and that’s a 1993 technology!

I’ve told people before – the real thing about E-mu’s technology is that it sounds much better than it should. That’s because I’m a jack-of-all-trades scientist/engineer. I designed not only the chips, but also much of the software architecture (though less in 90’s and beyond) and particularly the DSP algorithms.

I know when I design those algorithms that where their spec’s aren’t so good is where you don’t care. It’s too expensive to make a multi-channel instrument with perfect signal processing specs (and you’d rather have more channels that were good enough than fewer that were too good), but E-mu’s stuff sounded nearly perfect.

But we never turned this lead into a monetary advantage. One of my biggest mistakes was not to use crappy technology. When the Ensoniq Mirage came on the scene, it used “drop sample” pitch shifting – a technique I’d first seen in the PPG Wave (an early Emulator sort-of competitor) – I figured there was no way that serious musicians would use it, ’cause it sounds awful. Boy was I wrong, and Ensoniq sold more $ worth of Mirages in their first year that E-mu probably eved did of all Emulators combined. So “way ahead of the competition” doesn’t translate into monetary success.

What do you think was the basis of your success with the Emulator series?
Being on the right side of Moore’s law and being in the right place at the right time.
I heard a rumor that you at some point planned to release a SP-1600 sampling drum machine based on the Emax II technology. Can you tell me more about this? If it is true – why didn’t it make it into production?

I know we talked about an SP-1600, and it would probably have been based on the G-chip 1.0 technology, the same as in the original Proteus, and perhaps we would have included the H-chip technology, which first appeared in the EMAX II and later powered Morpheus.

The problem with any such instrument was return on investment. To do a proper drum machine, we’d need to write the software and design the hardware, both of which would cost money and perhaps more important, take resources away from other projects.

I’m sure we weighed the size of the market against these development and opportunity costs, and found the SP-1600 lacking. I don’t believe any real work was ever done on it.

Who did you consider your all-time strongest competitor?

OK, this is what I consider a “hard” question. I’ve always sort of liked Roland as a “good” competitor. By that I mean they mostly played by the rules, they did some very smart stuff, their competitive entries against ours were something I very much respected, and I felt like they respected us. In terms of companies from whom a competitive move nearly broke us, that happened at least twice.

The first was Sequential’s first drum machine, which abruptly dropped the Drumulator sales in a way we didn’t anticipate. Consequently, E-mu had a horrendous lay-off just before the E2 came out. The E2 saved the day, but the gutting of E-mu prevented us from capitalizing on that success; instead of growing we instead just rebuilt the damage from the lay-off.

The second was the Akai sampler that came out about the same time as the EIII. This was more our fault – we had designed a RAM SIMM socket into the EIII that Molex had discontinued because of reliability problems, but Molex never told us about the problems until we started seeing them in the field.

The EIII got a horrible reliability reputation, and the Akai was available competition, so again lay-offs and having to rebuild the company once we’d fixed the EIII and ultimately moved on to the EIIIx.

Certainly the “strongest” company that competes with us is Yamaha, but I don’t have much to say about them.

Why did you eventually choose to join Creative Technologies?

Around 1989, we realized that there was one significant problem with selling to musicians: your customers we always broke. Of course I say this mostly in jest, but there is some truth to it. While E-mu sold to the pros, there is no way to be a successful musical instrument company selling to the pros. They expect things for free. Most a very fine people, but business is business.

We got a big break when Stevie hugged the Emulator, but were very disappointed when the Emulator in the 2 page spread on Vangelis in Time Magazine had “E-mu Systems” covered in duct tape.
To be in a position to make a profit from pro musician instrument sales, you have to have a technology they can’t approximate elsewhere. And there is a huge marketing incentive for the Japanese giants to negotiate their products into the slots yours could take.

The bottom line – we made our money from the “semi-pro” musician. But they get their instruments from music stores – more negotiation against the Japanese giants for shelf space, tighter margins, etc. It’s just a tough business, and while E-mu stayed alive, we were never able to make enough money to create our own stability. The business was always up and down, and every product had to be a “home run.” Fortunately, they were.

So in ’98 we decided we needed “parentage.” That means we were looking for some deep pockets so we wouldn’t have to rebuild the company every time we made a mis-step. We thought of going public, but the Kurzweil IPO had made that very difficult.

So we looked for partners instead. As the “multimedia revolution” unfolded around us, we realized that the ideal partner would be Creative. After trying to get their attention for nearly a year, we received an impressive partnership offer from one of their competitors, contingent only on our never partnering with Creative. We knew Creative would be the stronger partner – we had two weeks to woo them.

Fortunately, we discovered that we had overlooked that both the COO and CFO of Creative Labs had E-mu gear in their home studios. When we got through to them, they were excited about a partnership, and arranged for the Singaporean founders to attend a chalk talk on our technology. I found out later that they decided after that chalk talk they were going to buy E-mu.

When the first proposed it, Scott and I were concerned that if they did buy E-mu, it would be the end of the business as we knew it. But the Creative folks said they wanted to continue E-mu, and they have been true to their word. As it’s turned out, the companies are remarkably compatible in culture as far as such things go. Sim, Creative’s CEO, still refers to E-mu as Creative’s “crown jewels.”

We had universal stock ownership at E-mu, so everybody who worked there got a good slug of money when we were acquired. A few (including Scott Wedge) were redundant and lost their jobs. But though Scott misses working here, I think he knows that we did the right thing. I’m sure that had we not joined Creative, there would be no E-mu today.

In designing the Emulator you came up with a way of compressing and expanding the samples. I am referring to the fact that it samples at 12bit – saves to disk at 8bit – and plays back the sample at something like 13-14bit (feel free correct me if I am wrong). This design was also used in the Emulator II series.
Can you explain what it does – and how you did it? Is it comparable to the ZIP or RAR technology used to compress files today – or is it something completely different?

I think you’re referring to the compression used in the E2. For those readers who are fluent in DSP, it’s adaptive D*PCM, implemented in the analog domain of the ADC and DAC. It is completely different from ZIP or RAR (which are far too complex to have been implemented at the time, and only poorly compress digital audio anyway – try it), nor is it related to MP3 or any of the perceptual CODECS of today (again, a realtime multichannel CODEC on these is hardly viable even today). It is probably most closely related to ADPCM, but works far better.

In the Emulator I and the Drumulator, we used Companding DACs (COMDACs) which operated on 8 bit data. But instead of giving each of the 256 states represented by 8 bits the same voltage span (as a normal “linear” DAC does), a COMDAC uses smaller voltage spans for the values near zero, and larger spans for the values far away.

The result is that while a linear DAC has its best distortion performance at full level, and for each 6dB the signal level drops the distortion increases by 6dB, a COMDAC has approximately equal distortion over 40 dB variation in level. COMDAC audio sounds a lot better that 8 bit linear audio, but it still sounds kinda lousy. That was the main limitation of the E1 and the Drumulator.

While researching the E2, we discovered that if instead of using the 8 bits of each sample to encode the actual voltage of the sound signal at that point, you used those 8 bits to encode the difference between the current sample and the previous sample, you got a significant improvement in distortion for many signals. That’s called differential PCM or DPCM. And if you used that same kind of varying voltage span to encode the difference (that is, used a COMDAC instead of a linear DAC to generate the difference), it sounded much better still.

That’s a form of Adaptive DPCM, but not the usual ADPCM that’s used today by Microsoft and others. And if you configure the feedback loop just right, you can sort of magically attenuate the quantization distortion at certain frequencies even more; that’s called D*PCM (pronounced “d star pcm”). The analog circuitry to do this is barely more complicated than a properly designed COMDAC or companding ADC.

This form of compression takes advantage of the fact that most audio has most of its energy in the lower frequencies. The distortion at low frequencies is extremely low – as you say its better than 12 bits, when below 1kHz at a 48kHz sample rate, and very low frequencies can approach 16 bit quality. And it turns out that most instruments that have lots of high frequency energy, things like cymbals, are pretty forgiving when it comes to distortion.

The only thing that sounds awful using this compression are pure, high frequency sounds. But who likes to listen to full level 10kHz sine waves – that hurts my ears. This is the sort of thing I meant above when I said that at E-mu, we specialize in stuff that sounds better than it should. We use what we know about the program material to make things sound best when they should, and if the specs are poor, they are poor in areas where you really don’t care all that much. The E2 encoding scheme was one of our first big steps where we implemented this design philosophy.

Do you feel that you were pushing technology back when you started designing samplers? Did you deliberately try to make the technology and components better than it was on the paper?

This is what I mean by easy questions – I think I already answered this a couple of times 😉

Yes – absolutely. I ALWAYS want to push the technology; that’s what’s fun, the more the better. And it’s those subtle things – not just how can I implement this DSP algorithm on a chip, but how can I improve it because I know it’s reproducing music? And what are the right trade-offs to give the musician the most bang for his design buck?

One thing Scott and I always talked about was the “balance” of our instruments. We didn’t want to give too many channels but too little memory. We didn’t want to have a sophisticated user interface on a modest instrument, or give too little control over a powerful engine. Or a memory so large the floppy disk couldn’t fill it. Or a memory so small that you could never take advantage of the optional hard disk. Users seldom think about these things, but they know a well tuned instrument when they play it.

Born in 1949, Dave Rossum grew up in San Jose, California. He was always interested in science, and won the grand prize in the San Francisco Bay Area Science Fair in 1964 with a gadget that looked at the spectra of high voltage sparks. Barely graduating from high school, he then received his B.S. in biology with honors from Caltech in 1970.

While doing graduate work at UCSC on the 3D structure of the ribosome, his thesis advisor foolishly introduced Dave to the university’s new Moog synthesizer. Shortly thereafter, Dave dropped out and founded E-mu Systems, which for over 35 years had been providing music synthesizers to professional and semi-pro musicians.

Dave holds three dozen patents; his innovations have changed the way music is made in the modern world. In 1993, E-mu was acquired by Creative Labs (makers of the Sound Blaster), for whom Dave has served as Chief Scientist for the last 12 years.

What he’s doing now:

That’s always hard to talk about. I’m doing some chip design work with Creative, and trying to contribute as much as time allows to E-mu’s products. At this point in my life, I leave the real work to the younger folks who are quicker with today’s tools. But I’ve architected some new DSP algorithms recently (nothing that will be earth-shaking, just more efficient), done a couple of novel analog circuits, and thought about how to do a decent music network (I’m a charter member of the “I hate MIDI” club).

One more thing

Ed Rundick, the “third founder” of E-mu, passed away last April after a couple of years battling cancer. I thought it would be appropriate, in the vein of “E-mu’s early years” to recount in one place the early history of Ed at E-mu.

In the fall of 1972, Scott and I had heard there was an ARP 2600 at Moyer’s Music in Santa Clara, so we went there in the hopes of learning something about the competition. We sort of snuck in and found the ARP in a corner. I began patching it up so I could measure stuff like contror rejection, frequency response, etc., using the synth as its own diagnostic signal generator.

After 5 minutes or so, the manager came over (busted!) and asked what we were doing. I figured I could snow him, so I explained that we were building a synthesizer, and the patch I had set up would tell me the control rejection of ARP’s lowpass filter.

Ed later described what went through his mind at that moment. He had tried himself to build a synthesizer, and found it impossible. But when he listened to what we were doing, he realized we really knew a lot about exactly how synths worked, and we might succeed!

Anyway, our kindred spirits quickly connected. Ed, who has always generous to a fault, gave us the ARP (it was his personal synth, not the music store’s) to take apart at our lab! (Interestingly, when I sanded down the VCO module and deduced the circuit from the circuit board traces, I realized that the VCO I had designed was way better than the ARP one anyway. That’s when we gave up looking at competitors circuits.)

From that moment, Ed was one of the “muons”. He visited us frequently, and we got him a job at Intel as a tech so he could learn more electronics. Ultimately, he was E-mu’s first employee, as well as the purchaser of our first analog modular synth system. Many of the modular schematics have Ed’s signature on them (he was draftsman, assembler, production manager, salesman, technician – you name it, Ed did it). He also had a great sense of humor (a requirement for working at E-mu in those days).

I remember one day after we had a “clean-up” of the lab, I found a box in which he had put all the “junk” that had no other home. It was filled with mis-matched screws and nuts, miscellaneous sample components, test circuit boards, and who knows what else. On the outside of the box was a label that read: “Caution – DO NOT OPEN UNLESS (1) You can’t find what you’re looking for anywhere else, (2) You REALLY NEED IT, (3) You are extremely patient person.”

In 1979, Sequential Circuits was enjoying the success of the Prophet 5, and was growing explosively. Meanwhile, the modular business at E-mu was waning, and we were about to start developing the Audity, based on the cash flow we were getting the Sequential’s royalty payments. So we encouraged Ed to go work at Sequential at a lead tech, helping get the Prophets out the door. Ed worked at Sequential for a year, but returned to E-mu in early 1980. He was the person who came up with the name “Emulator”.

Apart from me, Ed had the longest tenure of anybody at E-mu. Everybody knew him. He was very approachable; when Scott and I needed to know how the “troops” felt about something, we asked Ed.

He was E-mu’s conscience, and the ultimate muon. I sure miss him.