~ajxs/Yamaha DX7 Technical Analysis

Date: 2021.04.23

Location: ajxs.me

Foreword #

This article is intended to serve as an introductory technical analysis of the Yamaha DX7, detailing some of the known information about the synthesiser’s engineering. This article does not intend to be an exhaustive repository of such information, or as reference material for technical means. Instead it intends to provide an informative introduction to the subject.

The intended audience for this article is people with a computer science or similar engineering background who are interested in the technical aspects of synthesisers.

Acknowledgements #

The debt of gratitude this article owes to the hard work of others is so great that I would be remiss to even risk taking undue credit for their hard work.
This article would not have been possible without the work of Acreil, Raph Levien, the Dexed team, and Steffen Ohrendorf. I would also like to thank MadFame for his fantastic research into the DX7’s history, from which he has made an extremely informative short documentary.

Introduction #

In 1983 Yamaha Corporation released the now iconic DX7 synthesiser. Featuring a novel form of digital tone synthesis called frequency modulation, it would introduce musicians to a world of new timbral possibilities not possible with its analog contemporaries. By all accounts it was revolutionary, its brassy, metallic timbres would go on to define the characteristic sound of the decade. It would become the first commercially successful digital synthesiser, going on to sell over 200,000 units worldwide.

The frequency modulation synthesis that forms the basis of Yamaha’s FM synthesiser technology had its origins in the research of Stanford University professor John M. Chowning. Inspired by the groundbreaking work of Max Mathews at Bell Labs, Chowning shifted his course of study from music composition to focus on computer music. While experimenting with the modulation of a sound wave’s frequency, Chowning discovered that as the frequency of the modulating wave increased into the audible spectrum new and interesting harmonic partials were created in the carrier tone, while retaining the same pitch characteristics. Through a controlled application of this process a much wider range of tones could be synthesised than was possible using contemporary analog synthesisers.

The significance of Chowning's discovery was not lost on his colleagues, who urged him to patent this remarkable new technique. Interestingly, at this point in history it was not actually possible for an algorithm —or in fact any other kind of software— to be patented under United States law1, so Chowning enlisted the help of fellow academic Andy Moore from Stanford's Artificial Intelligence Lab to come up with a suitable hardware implementation. Their work together would yield the paper The Synthesis of Complex Audio Spectra by Means of Frequency Modulation[pdf], published by the Audio Engineering Society, and would ultimately become United States Patent No. 4,018,121.

In mathematical terms, FM synthesis is achieved by using the instantaneous amplitude of a signal (the modulator) to adjust the frequency of another signal (the carrier).
As convoluted as this sounds, it makes more sense when visualised:

There are many fantastic resources available online that give a much better explanation of FM synthesis than I can offer. Here is one particularly informative resource that I recommend. Another more mathematical approach to documenting frequency modulation can be found here, created by composer and synthesiser pioneer Barry Truax.

By the mid-1970s Stanford’s Office of technology and licensing had yet to successfully find a prospective licensee for their breakthrough in American music technology manufacturers2. At this point in time, the investment required for these companies to pivot towards the manufacture of digital synthesiser technology was simply too big a financial risk to take. The technological requirements were simply not commercially viable for companies specialising in analog equipment. To find a licensee, Stanford would need to look overseas. Despite not having a significant market presence on American shores, at that time the largest manufacturer of musical instruments in the world was Yamaha. As fate would have it, one of their senior engineers —an expert in digital technology named Kazukiyo Ishimura— was visiting their American branch at the time, apparently researching the design and development of the very kinds of integrated circuit technology that would make their future FM synthesis technology possible. He was willing to take a trip out to Stanford to hear Chowning’s breakthrough for himself. In Chowning’s own words: “In ten minutes he understood; he knew exactly what I was talking about” (Darter, 1985).

Yamaha would go on to purchase an exclusive license to the technology from Stanford, the rest is history. An extremely comprehensive and entertaining mini-documentary on the history of the Yamaha’s FM synthesiser range made by content-creator MadFame can be seen here.

In his memoirs, Roland Corporation’s founder Ikutarō Kakehashi wrote that he had also met with Chowning to discuss his research. Unfortunately for Roland, Stanford were unable to enter into any arrangement due to the exclusive license they had signed with Yamaha only six months earlier. Kakehashi admits that Yamaha were the right people for the job, given their capability to manufacture the LSI chips necessary to make the technology viable. (Kakehashi, 2002, pp. 194-195)

Roland’s technical R&D director Tadao Kikumoto —himself no stranger to innovation, having been the inventor of the iconic TB-303— spent a year poring over Yamaha’s technical literature looking for a way to implement an FM synth without running afoul of patent infringement3. By his own account he had also invested considerable time searching high and low for prior art to invalidate Yamaha's patent. This search would ultimately prove unfruitful (Reiffenstein, 2004, pp. 275-276). However unsuccessful these efforts may have been, Roland's research in the area of digital synthesis would lead to them produce their own groundbreaking digital synth: The D50.

Yamaha’s approach to patenting their FM technology is by all accounts comprehensive. It’s through these patents, service manuals, and the heroic reverse engineering efforts of some extremely talented engineers that we have been able to gleam what information we know of the DX7’s internals.

Hardware #

The beating heart of the DX7 is a Hitachi 63B03RP microprocessor, an enhanced version of Motorola’s venerable late 70s 6800 series of 8-bit CPUs built under license as a second-source supplier4. Clocked at around 1MHz, it provides the synthesiser a steady pulse, albeit not the kind of raw processing power required for real-time digital synthesis5. The DX7’s real capabilities lay inside two proprietary integrated circuits: the YM21280 FM-Operator Type S chip, otherwise known in Yamaha’s technical literature as the OPS, and it’s counterpart the YM21290 Envelope Generator or EGS. As their names suggest, the OPS and EGS are responsible for operator and envelope generation respectively.

Scanning of the DX7's analog input, such as keyboard, foot pedal, and panel switch events is handled by a separate “sub-CPU”, a Hitachi 6805S 8-bit processor. The sub-CPU communicates with the synth's analog peripherals via a multiplexed A/D converter, and with the main CPU via a complex handshake mechanism described in detail in the service manual.

MIDI I/O is handled by the main CPU, via an integrated UART interface. Referred to in the service manual as an ACIA (Asynchronous Communications Interface Adapter), and as an SCI (Serial Communications Interface) in Hitachi's technical literature.

We know from the service manual that the OPS processes all 16 six-operator voices in ~20.368032μs. From this we can deduce that the synth’s overall sample rate is 49096Hz.

The DX7's smaller desktop module cousin, the TX7, is built around an entirely different architecture. Featuring a different microprocessor, the Hitachi HD63A03X. It does not include a sub-CPU, with all input scanning being performed by the main processor. The TX7's service manual —an incredibly detailed source of technical information— provides a memory map for the CPU, as well as a flowchart of the ACIA interrupt routine, timer interrupt routine, and main executive loop.

YM21280 #

Each of the DX7’s six operators consist of a digital sine wave oscillator, with its own independent envelope. In technical terms, these are what are known as numerically controlled oscillators: A phase accumulator register is used to store an index into a lookup table of wave amplitude values stored in ROM. In this system the angular velocity of the oscillator’s wave is controlled by incrementing the phase accumulator each sampling iteration with an amount proportional to the desired frequency6. The result of this process is a sine wave output supporting smooth modulation of its frequency.

We know from the patent literature that the YM21280’s pre-calculated sine wave lookup table is stored in logarithmic format. Two studious engineers Matthew Gambrell and Olli Niemitalo were able to decapsulate the YMF262 (otherwise known as the OPL3, a nineties-era 2-operator Yamaha FM sound-chip designed for use within PC sound-cards), and read the binary ROM data encoded in the chip’s die. Thanks to their incredible work we now know the format and content of this sine wave table.

Taking full advantage of the sine wave’s symmetrical property, the YMF262 stores only a quarter of the sine’s full period. Flipping and inverting the data to assemble the full wave as needed. It’s highly likely that the YM21280’s sine wave table is formatted similarly7.

Using a novel mathematical trick known to anyone familiar with a slide-rule, storing the oscillator’s wave data in logarithmic format allows for scaling the amplitude of the wave without the need for multiplication8. Given the logarithmic identity: log(xy) = log(x) + log(y), storing the operator’s wave and envelope data in logarithmic form allows for replacing the multiplication of the wave’s instantaneous amplitude with the envelope gain coefficient with a relatively cheap addition operation.

We know that the YMF262 stores a lookup table of exponential values which is used to convert the product of this calculation from logarithmic to linear format, taking advantage of the exponential identity 2log2(x) = x. The service manual of the CE-20 (an earlier, lesser-known FM synthesiser produced by Yamaha) confirms the presence of such an exponential table there too. Although the DX7’s patent literature does not explicitly mention such a thing, it is reasonable to assume this is how the logarithm-linear converter described in the patent functions.

The best description I’ve found of the process of constructing the operator samples from the sine and exponential tables has been made by the amazing Steffen Ohrendorf, who has documented the process for the OPL3 here in a brilliant, mathematically correct notation.

Once the correct phase data has been calculated for each of the operators, the actual frequency modulation is performed. The sinusoidal output of one operator (described in the patent as f(ωmt)) is then used to modulate the phase angle9 (Kωt) of another operator acting as a carrier. In the patent’s poetic flourish: “the sine wave table provides an instantaneous amplitude value of the frequency modulated signal, i.e., sin {Kωt + f(ωmt)}”. The service manual provides a surprisingly in-depth description (considering that no datasheet for the EGS or OPS is publicly available) of how data is stored in the OPS’ internal registers during the synthesis process. Describing the process used in several different distinct algorithms in detail. This process is not worth paraphrasing here. I would encourage interested readers to refer to this document.

Thanks to the brilliant research of synthesiser enthusiast Acreil, we have evidence that the length of the DX7’s full sine table likely consists of 4096 entries: He makes the remarkable observation that when using a modulator frequency much lower than the carrier, harmonics appear in the wave corresponding to the length of sine table. With a modulator frequency of f, and a hypothetical full sine period of n samples, images of the harmonic appear at n*f. In the case of the DX7, n=4096. Interestingly, these harmonics become more attenuated as the sine table length increases. This factor likely contributes heavily to the trademark gritty digital quality of the later FM chips.

Unfortunately, we can’t infer from this any additional information regarding the bit-depth of the DX7’s sine table. The bit-depth of the sine table in the YM3812 chip appears to be 12-bit, so it’s reasonable to infer that the DX7 is similar. The DX7 service manual supports this hypothesis, showing that amplitude data is transmitted from the YM21280 to the DX7’s DAC over a 12-bit parallel interface.

Feedback

The DX7’s novel feedback mechanism is implemented by storing the result of processing an operator and using it to modulate itself over multiple iterations. In technical terms, the instantaneous output amplitude of the last operator processing iteration is added to the phase accumulator value used as the input for the current. This feedback input added to the phase accumulator is scaled according to the feedback level parameter. As the feedback iterations increase the sinusoidal operator output approaches the appearance of a sawtooth wave. For interested readers, US patent 4,249,447 contains a detailed description of the feedback synthesis process.

The ‘anti-hunting’ filter mentioned in the patent literature10 is applied as an anti-aliasing measure, implemented by averaging the latest sample with the last. This page, written by composer, and researcher Risto Holopainen, gives a very detailed overview of the hunting phenomenon described in the patent by Yamaha engineer Norio Tomisawa.

YM21290 #

The YM21290 (EGS) is the entry point of the DX7’s synthesis process. It receives data from the CPU, and generates the frequency data and envelope amplitudes. This data is then sent to the YM21280 over a parallel interface to process the operators and generate the actual tones.

We know from the service manual, and schematics that the frequency data (Kωt) has a word length of 14-bits. Similarly, we can see that the amplitude data has a word length of 12-bits.

The patent describes the phase generator as having a “96 stage/22-bit shift register”. Without evidence to the contrary it is reasonable to assume that the patent is correct, and that the width of the phase accumulator register is 22-bits.

According to Raph Levien, whose fantastic engineering serves as the basis of the highly acclaimed Dexed emulator: “Through measurement, it's clear that the 12-bit envelope value is a simple Q8 fixed-point representation of logarithmic (base 2) gain. Linear gain is equal to 2^(value / 256). The steps are particularly clearly seen in plots of amplitude for slow-decaying envelopes. This gives a total of about 96dB of dynamic range, in steps of approximately 6 / 256 = .0234 dB, which is smooth to the ear”.

The exact internal representation of this numerical data remains the subject of debate11, however Raph Levien’s conjecture above regarding the representation of numerical data stands up to scrutiny. These formats have been used in the emulation of the DX7 Mk1 engine in Dexed. They say the proof is in the pudding: in this regard the accuracy of Dexed’s emulation speaks for itself.

Phase Generator

The DX7 patent describes in light detail the functionality of the phase generator, a sub-component of the EGS used to create the phase angle data (Kωt), according to what key is pressed. In patent’s florid verbiage: “The adder 14 is provided for changing the value of the key code KC in response to the coefficient k. A frequency number generator 15 generates, responsive to the output of the adder 14, numerical data representing an amount of phase change per unit time, i.e., a frequency number.” This likely refers to the addition of the key code (likely represented in cents/100) together with the scale of an operator. An internal exponential table is most likely referenced to convert the resulting logarithmic value into hz, from which the phase increment can be obtained. This is consistent with Figure 1 in the patent, which shows the KEYCODE (KC) and the output of the OPERATOR CONTROL DATA REGISTER (k) being applied as inputs to an adder, then to the FREQUENCY NUMBER GENERATOR, ultimately yielding Kωt. The patent describes this component as: “A frequency number generator 15 is constituted of a logarithm-linear converter converting the logarithmically expressed frequency number to the linearly expressed frequency number kω.”

Generally speaking, the formula for obtaining the phase increment for a desired frequency in such a system is:

The exact method by which the YM21290 calculates the phase increment for a given key is unknown, and without further experimentation (see afterword), it will likely remain so. We can gain some insight into likely methodologies by examining Yamaha’s later FM chips, Many of which were mass-produced and used in various different electronic appliances such as PC sound-cards and arcade game consoles. As a result, much more documentation is publicly available.

The YM2151 sound chip was Yamaha’s first single-chip FM synth system, and the direct successor to the YM21280. Released in 1984 —hot on the heels of the DX7— it contains only four FM operators and does not require a separate envelope generator chip. It is used in Yamaha’s four operator FM synthesiser range: the DX21, DX27, and DX100. While its design is clearly distinct to that of the chips used in the DX7, it’s highly likely that they share implementation details.

The YM2151 application manual describes in detail the internal registers involved in the calculation of the phase increment, however it stops short of providing an exact formula. I won’t describe its novel system of phase generation here, however anyone interested should refer to the publicly available documentation for the OPM, OPL, and OPL2 chips. One particularly good resource on programming these 4-operator FM chips can be found here.

As the OPS section above states, the envelope amplitude is stored in logarithmic format, which allows for the scaling of the wave amplitude to be performed in a relatively inexpensive manner. The service manual contains an extremely informative block diagram of the EGS showing the process of assembling the final envelope amplitude value. It demonstrates that the rate scaling and envelope modulation values for each operator are added to the final envelope amplitude value, heavily implying that these too are stored in logarithmic format. Similarly it shows that the detune and pitch modulation buffers are added to the phase increment value before it is transferred to the OPS.

DAC #

The DX7 uses a time-multiplexed DAC with two sample and hold circuits. This means that the synth’s 16 voices are output in two repeating time slices. This is done for purposes of maximising the dynamic range of the synth’s output and limiting intermodulation distortion between the different voices. This can be seen by analysing the synthesiser's output in an oscilloscope, and is confirmed by the TX7 Service Manual.

The actual digital-to-analog conversion of the synthesiser's output is performed by a BA9221 12-bit bipolar DAC chip with a parallel data interface. Interestingly, the reference voltage of the DAC is not hard-wired, but is controlled via a DAC circuit consisting of a resistor ladder connected to an analog multiplexer (This chip is labelled as IC46 in the schematics). This effectively allows for 8 discrete volume levels controllable via software. This circuit is connected to the main CPU's data bus and is presumably controlled via the MIDI amplitude parameters. This particular setup seems to vary widely between different synth implementations, with the TX7 and TX802 having different setups entirely.

The most interesting feature of the DX7's DAC is that the output of the aforementioned BA9221 is fed into another analog multiplexer (labelled as IC41 in the schematics), which functions as an exponential scaler for the DAC output, effectively expanding the resolution of the DAC to 15-bits. This analog multiplexer consists of a TC4066BP quad bilateral switch connected to an R-2R resistor ladder. This is described as the amplitude scale factor in the service manual, and is controlled via the SF0 ~ SF3 pins on the YM21280. This effectively converts the 12-bit linear DAC into what could be described as a 15-bit floating point DAC, having a 12-bit mantissa and 3-bit exponent. The TX7's service manual describes this functionality of this circuit in the following manner (corrections mine):

"The output data of the OPS is 12-bit. However, to make this the equivalent of 14-bit [sic], the lower levels are expanded 2, 4, and 8 times respectively. To return this to the original value, shift data (SF0~SF3) is sent out.

The-12 bit digital data from the OPS is sent to the DAC IC24 and converted into an analog signal. This 12 bit digital data has been expanded inside the OPS, so the IC26 and the connected resistances will return it to the original level. This is controlled by the shift data sent from the OPS (SF0 ~ SF3), which is sent at the same time as the 12 bit digital data.
The shift data is as follows:
When the data sent to the DAC has been shifted 1 time, SF0 sends "High".
When the data sent to the DAC has been shifted 2 times, SF1 sends "High".
When the data sent to the DAC has been shifted 4 times, SF2 sends "High".
When the data sent to the DAC has been shifted 8 times, SF3 Sends "High"."

Misleading terminology aside, this confirms the functionality of the Scale Factor pins and this DAC circuit. The TX7's schematics refer to this as a compander, which is analogous to the role this circuit plays within the synthesiser. It was highly likely that the reasoning behind this curious setup was simply the unit cost associated with higher-quality DAC chips. Yamaha's engineers likely did not have access to reasonably priced DAC chips capable of the required analog performance, necessitating some clever engineering on their behalf12.

After the DAC stage, an analog amplifier circuit is used to control the final output volume. A lowpass filter with a fixed cutoff frequency of 16KHz is then applied to remove high-frequency aliasing.

Afterword and Going Further #

This article serves as a cursory technical analysis of Yamaha’s DX7 synthesiser. The technical subject matter underpinning Yamaha's DX series of synthesisers is a topic easily as broad as it is deep. There are several areas that I have regretfully had to neglect in my coverage for the sake of brevity or the difficulty of research, such as the keyboard scanning implementation for just one example.

I have done my best to ensure the factual accuracy of the information presented here, however I permit the possibility that some inaccuracies may be present. If you have any additional relevant information, or can spot any errors in the article, please email me and let me know. You’ll be credited in the article for your efforts!

The efforts of developers working on the various emulations of the DX7, particularly the Dexed emulator, represent what is likely the most accurate understanding of the DX7’s internal implementation available.

To get a more information on the DX7’s implementation, some next steps could include:

Decapsulate the YM21280 and YM21290: It’s likely that this has been performed by Yamaha’s competitors, if not hobbyist reverse engineers, at some point in the past. However there does not seem to be any public account of this. Decapsulating the IC’s would reveal at minimum the contents of the sine and exponential tables, and potentially provide further details as to the internal representation of the numeric data.

Acreil points out that several of Yamaha's digital pianos, such as the PF80 or the Clavinova CLP-30, feature the YM2604 OPS2 and YM2603 EGS2 chips. These chips are likely very similar in nature to the YM21280/YM21290 and would provide good alternative candidates for IC decapsulation. The DX9 also uses the same chips as the DX7, and can potentially be obtained more cheaply than the DX7.

Probe the circuitry: Additional information regarding the format of data transferred between the ICs could be gained by an engineer using a logic analyser on the lines of the various internal chips. For one example, probing the input lines to the YM21290 could potentially yield additional information regarding the operation of the chip’s phase generator.

Disassemble the OS ROM: Additional information regarding the transmission of data between the key scanner, CPU(s) and the YM21290 could be obtained by disassembling the OS ROM. As daunting a task as this may seem, it is not outside the realm of possibility. It has likely been performed in the past13, however no online record can be found. The service manual even lists the address space layout for the CPU, which would be invaluable in reverse engineering the operating system ROM image. If one were to disassemble the ROM, it would theoretically be possible to alter it to dump debugging information out via the MIDI output port, recompile and flash it to an EPROM chip.

Obtain original documentation: Obtaining any original documentation from Yamaha would certainly settle the matter of reverse engineering the DX7 definitively. Obtaining datasheets or programming manuals for the OPS or EGS chips would likely provide useful information on their internal register layout and implementation. Unfortunately I am not optimistic about Yamaha ever releasing this information, were it available. Corporations tend to have narrow views about the sharing of proprietary information. For what it’s worth, I emailed Yamaha enquiring about the possibility of obtaining this documentation and was told by their contact centre they have no access to this documentation.
It’s also likely that such documentation may have been lost in the intervening decades, and that if this documentation were to still exist it may not have been translated into English.

When it comes to telling the rich history of technical innovation that led to the Yamaha DX7, this article does not even scratch the surface. To fully cover not only the technical details, but the history, would take easily a year of research and an entire series of articles. Despite the limited scope of this article, I hope it has served to entertain and inform readers. I hope that it can be of some practical use in future reverse engineering efforts. If you are involved in any such efforts and this article was useful, I’d love to hear about your work.

References #


  1. The history of software patenting is interesting topic in and of itself: https://bitlaw.com/software-patent/history.html. This legal curiosity accounts for the particular format of Chowning's original patent, which focuses on a 'preferred embodiment' represented in the form of an integrated circuit. Consisting of a series of gates.
  2. These were organ manufacturers, such as Allen, Hammond and Lowery. By Chowning’s account, their engineers were either inexperienced in digital systems, or outrightly disinterested (Darter, 1985). This shouldn’t surprise readers. It was the mid-seventies, by all accounts the heyday of the analog synthesiser. Something the DX7 was going to change.
  3. The historical record attests to Yamaha and Stanford having sent patent infringement notices to various companies who had implemented FM synthesis in their devices. It is probably for this reason that synthesiser manufacturers throughout the 80s and 90s implemented all manner of cross-modulation mechanisms into their synthesisers, calling them anything other than 'frequency modulation'.
    Dave Smith offers an interesting account of the patenting strategy employed by the large Japanese firms, contending that it was a defensive tactic against spurious claims of infringement: "If you have a lot of patents and somebody comes after you, chances are one of your patents will overlap enough with one of their patents that you can negotiate a deal so nobody gets hurt. Whereas if you don't have anything to offer and you have nothing in your stable of patents, then you're stuck" : (Reiffenstein, 2004, pp. 247-248)
  4. It is commonplace in the electronics manufacturing industry to not utilise any components for which only one supplier is available. Where a case of supply-side disruption could jeopardise manufacturing. As a way to ensure that their components weren’t passed over in favour of safer bets, companies licensed out their IP to ‘second-source suppliers’ to manufacture compatible components. AMD, possibly the most famous example, made their mark this way.
  5. It should be mentioned that the 6809/6309 was an extremely capable processor in its time, one of the last mainstream 8-bit processors. As a result it contains many advanced features not found among its contemporaries, such as the Direct Page Register. A special register used to store an 8-bit base address, which allowed the use of instructions addressing absolute 16-bit address offsets stored within the machine’s 8-bit accumulator registers. A novel feature for its time. However capable the processor was, at these low clock speeds it was simply not capable of meeting the arithmetic requirements of 16 voices of 6-operator FM synthesis.
  6. This use of a Phase Accumulator is standard practice in Direct Digital Synthesis, used in a variety of different engineering disciplines.
  7. As of the time of writing, I’m unaware of anyone who has decapsulated a YM21280. With the going price of a DX7 on second-hand marketplaces being what they are, I’m certainly not about to volunteer my own for the task.
  8. A comparatively expensive operation relative to integer addition, in terms of both clock size on the CPU and die size on the LSI chip. Since the DX7’s DAC is clocked in at an impressive 49096Hz, with 16 6-operator voices -not to mention feedback operations- each featuring it’s own independent gain and envelope, this trick avoids the need for millions of multiplication operations per second.
  9. Whether the DX7 uses frequency modulation or some other form of phase modulation (or whether FM is even classifiable as phase modulation) has been the source of some serious contention. Tom Wiltshire from Electric Druid does an excellent write-up of Casio’s similar ‘phase-distortion synthesis’ technology here.
  10. The patent describes this filter mechanism as: “Such averaging operation serves to the prevent the hunting phenomenon in the circulating type frequency modulation operation”. An illustration of this phenomena can be found in the patent US4249447.
  11. Carbon14 makes a detailed analysis of the OPL2 here, and credibly argues that the internal representation of the phase accumulator could not be floating point based upon the clamping observed during analysis of operator summation.
  12. Acreil points out that the increased quantization distortion resulting from this setup is neglibile and is likely not responsible for the DX7's trademark distortion, noting that the source is more likely to be poor resistor matching in the compander's resistor ladder, or even the settling time of the BA9221 DAC being insufficient for the high frequency at which it is operated in the DX7.
  13. Aftermarket operating systems for various contemporary synthesisers existed in this era, distributed via EPROM chips. One particularly good example is the Soundprocess OS for the Ensoniq Mirage, developed by Mark Cecys.