Digital Output for SY77/SY99

The Yamaha SY99 is a synthesiser combining frequency modulation synthesis (branded as Advanced FM) and sample-based synthesis (branded as Advanced Wave Memory 2) and the direct successor to Yamaha's SY77/TG77

Moderators: parametric, Derek, Saul, Fozzer

JK1974
Member
Member
Posts: 85
Joined: Tue Oct 02, 2012 1:01 am

Digital Output for SY77/SY99

Unread post by JK1974 » Sat Oct 20, 2012 12:19 am

Hi,

when I started working with my SY99 20 years ago, I was on a trip hating every analogue noise. As it was (and still is) just a hobby to me, I did not invest in really expensive studio equipment, apart from a
Terratec EWS64XL soundcard with a value of >400€ with a working, (near) bit-identical S/PDIF input (compared to the 48 kHz crap from Creative Labs at this time). And I had a DAT recorder that allowed me some kind of high quality AD conversion without disturbing noise. The recordings were transfered to the PC using the S/PDIF input of the EWS64XL- less noise than using the PC itself for recording.
I haven´t used the SY99 for years now, I turned it on again around three weeks ago. But all the time, having the SY99 next to me in my private office, I remembered that I have read about an S/PDIF extension for the SY99. Searching the web, I got no proof that this really existed.

After having become interest in the SY99 again, I searched the documents I collected 10-20 years ago - and I found the advertisment, and I remembered having brought it from the Frankfurt Music Fair, apparently in 1998.
Image

The (german) text talks about an AES/EBU interface for the SY77/SY99, the company also had an AES/EBU to S/PDIF converter. You could even use a frequency of 44.1 kHz (instead of 48 kHz), but then you seemed to have to transpose the whole sound. And there was also an interface for the TG77. The company´s name was Audio Service, located in Hamburg, and searching the web again yesterday, I found the company and at least a PDF about the TG77 Digital Interface in the download section of http://www.audio-service.com.
The company does not seem to offer this hardware extension anymore, and in the catalogue of 1998, they talked about 950 DM (around 475 Euros) for this modification. Even today, I would not spend such amount of money for the digital extension - I would rather look for an alternative... ;)

I am not too much in electronics. However, two years ago, I started taking a look into the open Arduino microcontroller project (http://www.arduino.cc) and electronic basics. With this basic knowledge, I took a look at the schematics of the SY99, searching for a place where a digital signal could be taken from the SY99. I seem to have found it.

On page 13 of the documentation, on the left upper part, there is a Yamaha specific chip called "AFD0 IC206". This one gets information from a chip described as "MIX5 IC228". Searching the web, it seems that there is nearly nothing known about the "AFD0 IC206". The documentation claims, that it is an DA converter chip.
But, looking up and right to it, there are know components: On the right, you find amplifying chips - something that means, that the signal must have been DA converted before. And on top, there is a chip called "PCM56P-Y" that also sends information back to the "AFD0" as is called "Digital Analog Converted" on page 30.

In fact, this PCM56P is a DA converter chip by the well-known manufacturer BurrBrown. Although I don´t understand enough from it´s data sheet, the digital signal seems to be feed to the pins 5-7 (CK, LEC, DATA), and the used protocol looks similar to the I2S protocol (compare the datasheet to the timing diagram at http://de.wikipedia.org/wiki/I%C2%B2S; the english version does not contain this diagram). A (minimal) conversion seems to be necessary to make it I2S compatible - I don´t know, if this could be done by a cheap logic chip or if a microcontroller would be necessary. For the conversion from I2S to S/PDIF, there are chips available.

However, I have no idea about the voltages used (and how to power the additional chips), but I think, as we are still in the digital domain, at least the output and input voltages must be equal and -in theory - connecting additional chips should be possible without problems. But I don´t have an oscillator to check the findings, nor do I have a spare SY99 to test with. ;)

(Of course, no 44.1 kHz would be possible, and I currently don´t see a solution to use it as S/PDIF slave. And the conversion to higher sample rates would be an another challenge. But this basic digital output would be enough for me as a first step.)

Any comments?



User avatar
Clyde
Global Moderator
Global Moderator
Posts: 6073
Joined: Mon Feb 16, 2004 1:00 am
United States of America

Re: Digital Output for SY77/SY99

Unread post by Clyde » Sat Oct 20, 2012 3:25 am

Interesting post, a digital out for the SY99 would be a great enhancement. And the AES/EBU output conversion would have been an easy starting point, alas no longer available just like the Musitronics 8mb ram expansion for the SY99 is no longer available. If any new upgrades for this great synth come about, it will most likely be from a dedicated user/users with electronics savy and a "can do" attitude that will bring it about. Thank you for posting this info!
Clyde


DX7IIFD, SY77, SY99, Hammond C3, Steinway L, CP300, AW1600, etc.

fingers109
Posts: 1
Joined: Tue Mar 19, 2013 9:52 pm

Re: Digital Output for SY77/SY99

Unread post by fingers109 » Wed Apr 24, 2013 3:46 pm

I've just started looking into this! So, for anyone whose interested here are my conclusions from looking at the circuit diagram and the PCM56P datasheet:-

The YM3029 sends digital data to the PCM56P 16bit for each of the 4 channels. The PCM56P sends back analogue to the YM3029 which then sends the correct analogue signal to its analogue output (& amplifiers).
So you only need 1 DAC for all four analogue channels. (the PCM56P datasheet as a 2 channel example of this)
So, the YM3029 controls the timings for each channel DAC conversion.
In theory, all you need to do is take the digital output from then YM3029, DAB(digital signal), LE (latch) and CLK and split them into 2 s/pdif channels. Then clock this with your external word source and bobs your banana.....
simple heh???
Odds on me getting any further than this is the next 20 years is slim....
Help us sector101.co.uk you're our only hope.....



User avatar
tux
Senior Member
Senior Member
Posts: 1081
Joined: Wed Oct 10, 2012 1:42 am
Kiribati

Re: Digital Output for SY77/SY99

Unread post by tux » Wed Apr 24, 2013 6:19 pm

fingers109 wrote:Help us sector101.co.uk you're our only hope.....
I think Brian might soon start muttering something about "give them a hand and they'll take an arm..." :wink: :mrgreen:


My Yamaha RM50 page
My Yamaha synths: RM50, TG77, TG500, EX5R, CS6R (with PLG150-AN and PLG150-DX)

JK1974
Member
Member
Posts: 85
Joined: Tue Oct 02, 2012 1:01 am

Re: Digital Output for SY77/SY99

Unread post by JK1974 » Tue Aug 13, 2013 3:09 pm

Searching for "how to convert PCM to S/PDIF", I found the CS8406 by Cirrus Logic (http://www.cirrus.com/en/products/cs840 ... Key=CS8406; forwarded by http://www.alpha-ii.com/Info/snes-spdif.html). Seems to be the right piece for this issue, but I still not have become an electronics engineer.



db7

Re: Digital Output for SY77/SY99

Unread post by db7 » Tue Aug 13, 2013 4:55 pm

I presume it can function at sampling rates other than the very excessive 192 kHz in order to be useful here? I’d guess so, and I ask, because a very common rate for audio sent over S/PDIF is 48 kHz, which I think may be very much in our favour for the following reasons.

The only rate cited in the service manuals of the 77/99 is 48 kHz, which is the maximal sampling rate of AWM waveforms. It is logical to assume the AWM chip, M3, always runs at this rate, interpolating lower rates up. Note that there are two M3s: A and B, much as there are two 8-voice pairs of OP(erator)S3 and EGM2: 1 and 2. The documentation makes it clear that the output of the OPS3s is sent via the M3, even if RCM is not being used. Thus, it seems logical to conclude this raw, pre-panning/-DSP stage of mixing AFM and AWM operates at 48 kHz.

The DSP stage(s)—two pairs of separate modulation and delay/reverb chips in the 77, one pair of catch-all DSPs in the 99—may well upsample during their internal processing, and they just might output at a rate other than 48 kHz, although I doubt the latter.

So, it seems likely that the final signal going into the DAC (AFD0/YM3029) is sampled at 48 kHz. The rest is ‘just’ a matter of determining the precise electronic format in which that signal is conveyed, hijacking it, and converting it to a modern standard for transmission. Be warned that the interception and reformatting might require a custom-programmed IC such as an Arduino: there might not be any ready-made one out there that can handle whatever format the AFD0 takes as input.

However! fingers109 claims there’s an intervening analogue stage between two digital ones in the AFD0. Is it possible this acts as an analogue exponent (read: amplifier/attenuator) of a digitally converted mantissa? That happens in that classic old DAC, the YM3012, and might be supported by the mentions in the service manual of pins for “exponent” and “Volume level” on the AFD0. I can’t yet contort my brain enough to understand the flow of data in the 77/99*, but all of this makes me wonder… If there is an intermediate analogue stage that acts as an exponent, we’d need to tap its input and simulate it in the digital realm, making it (ahem) exponentially more likely that we’d need a bespoke IC to convert to S/PDIF.

Of course, I’d welcome insight from those who might be able to decipher further what’s really happening in there. And, needless to say, I’ll be very interested if those with the requisite electronic knowledge can develop anything towards the goal of a functional digital output.

* The circuitry in the 77/99 is far more complex than Yamaha’s previous synths, indicating just how much of a jump in technology these were.



JK1974
Member
Member
Posts: 85
Joined: Tue Oct 02, 2012 1:01 am

Re: Digital Output for SY77/SY99

Unread post by JK1974 » Tue Aug 13, 2013 7:41 pm

The mentioned Cirrus Logic chip supports all sampling rates, 48 kHz included.

And the digital output format of the YM3029 is known: Just have a look at the PCM56 datasheet - it describes how the input format at pins 5-7 has to be formatted like. ;)
If I have understood correctly, the CS8406 can be set to different input protocols, so we need to know if the YM3029 "format" is also supported directly or at least can be programmed to support it. Basically, an Arduino is not fast enough to take the YM3029 digital audio data directly, but it could/has to be used to set the right parameters of/initialize the CS8406 chip.

I also have no idea about the analog part of the YM3029. Even if it is a kind of "general purpose" chip e.g. to control the PCM56, I also don´t understand why they don´t simply connect the analog outputs of the PCM56s to the S/H LPF chips.
Could it be that the YM3029 includes some kind of EQ to give the SY77/SY99 a "certain kind of sound"? 20 years ago, I had the feeling of a lack of "brillance", and so I often used a EQ to boost the higher frequencies on my own samples before I transferred them to the SY99.

BTW.: Does anyone have an idea why there are the S/H LPF chips after the YM3029 at all? I thought, LPFs (low pass filters?) are included in general in DACs to avoid aliasing.



db7

Re: Digital Output for SY77/SY99

Unread post by db7 » Tue Aug 13, 2013 10:56 pm

I never got around to checking the datasheet for the PCM56. Thanks for the summary. I guess the only uncertainty now is what that added analogue stage is doing. I still suspect it could be something like the split in the YM3012 between mantissa, which the DAC itself converts down, and exponent, which it sends to an analogue buffer amplifier whose output then attenuates the converted mantissa. But maybe I’m wrong. I know next to nothing about the designs of different DACs, so maybe an analogue exponent is an unusual feature or something that’s not advisable except in relatively old/simplistic DACs like the YM3012.
JK1974 wrote:Could it be that the YM3029 includes some kind of EQ to give the SY77/SY99 a "certain kind of sound"? 20 years ago, I had the feeling of a lack of "brillance", and so I often used a EQ to boost the higher frequencies on my own samples before I transferred them to the SY99.
I realy have to doubt the DAC or any other digital IC has any kind of non-configurable filter built-in. As for the analogue side, although I have no proof until I can decipher the schematic enough to quantify its filters, I really doubt it too. I never notice any lack of high frequencies or whatever you were describing there. Sorry. :P
JK1974 wrote:BTW.: Does anyone have an idea why there are the S/H LPF chips after the YM3029 at all? I thought, LPFs (low pass filters?) are included in general in DACs to avoid aliasing.
Using a low-pass filter after a DAC is absolutely industry-standard basic practise. Here, the LPF functions as a reconstruction filter to attenuate away unwanted shifted copies of the desired signal that start appearing immediately at sampling rate ÷ 2 (the Nyquist frequency).

Images are generated because a basic property of the zero-order-hold (stair-stepped) waveforms coming out of a DAC is they additionally contain mirror-images of the desired spectral content around integer multiples of the sampling rate. Note I said zero-order hold. Sample/hold is an inaccurate but very overused term here: it should be reserved for analogue input being sampled digitally, not the converse.

This picture shows how the images are distributed, as well as indicating why any sane DAC nowadays, like those playing CDs at a nominal sampling rate of 44.1 kHz, performs oversampling to push them away, thus simplifying the design of post-DAC filters and reducing their unwanted effects upon phase, etc.:
Image

The reconstruction filter will attenuate high frequencies, to a degree that depends upon its construction. Visually, this smooths off the stairsteps, creating a waveform that looks nicer, curvier, whatever. In terms of sound, perhaps no difference would be noticed as the images would be well outwith human hearing. However, it is good practise to remove such high frequencies in case they risk damage to speakers, or modulate with audible frequencies causing intermodulation distortion in the audible range.

Imaging does get called aliasing a lot, but I find that introduces too much confusion with the other meaning of that term, so I try to avoid the ambiguity.

The other, more musically common definition of the word aliasing is when a source, be that a digital oscillator or an analogue input to an ADC (not DAC), contains frequencies equal to or greater than the Nyquist limit. Such exceeding frequencies become reflected, around the axis of the Nyquist frequency, into the audible range as unwanted components that introduce unwanted tones, non-harmonic distortion, etc. Where the signal in question is an analogue input, an LPF should be placed before the DAC [edit] whoops: ADC [/edit] to filter out frequencies beyond Nyquist that would cause aliasing; this is an anti-aliasing filter.

In neither of these cases is the LPF built into the DAC or the ADC. These devices need only fulfil their stated purpose by sampling in or out n times per second. Any filtering is usually left to separate analogue circuitry. There are digital filters that may be used in more advanced/modern devices, although even those would almost certainly be separate chips and not integrated into the converter itself.



JK1974
Member
Member
Posts: 85
Joined: Tue Oct 02, 2012 1:01 am

Re: Digital Output for SY77/SY99

Unread post by JK1974 » Wed Aug 14, 2013 10:40 am

Thanks for your clarification. I knew that a LPF is needed but really thought that it is part of a DAC because it is always needed.

Concerning my EQ idea: I believe to have generated a sine sweep sample with CoolEdit in the old days, transferred it to the SY99, re-recorded it and found an attenuation at the higher frequencies. In general, I missed the brilliance of drumloops I sampled at that time (compared to the "original"), and sounds like a K1 Piano or Roland Pizzicato did not have the assertiveness that the "original" had when played through the SY99.
Furthermore, in reviews of synths of these days in Germany, they always had judgements of the overall sound of synths, and different synths seemed to have different sound characteristics. Assertiveness, the sound of low and high frequencies were often mentioned.

What about a (slight) dynamic compression? Other ideas what could be done in the analog domain to clean/optimize the sound?
Is there a volume setting somewhere within SY99 independent of the volume sliders where you change the volume of output 1-4?
Does the PCM56 output the right levels or is some kind of amplification needed in general to fullfill studio level standards?



mjbrands
Member
Member
Posts: 15
Joined: Sun Aug 25, 2013 3:49 pm

Re: Digital Output for SY77/SY99

Unread post by mjbrands » Mon Sep 09, 2013 10:37 pm

I've had a look at this and wonder if it is worth it.

I haven't had my board open yet, but to me it looks like the PCM56P DAC (which is mono and 16 bit) is running at 4x the audio frequency (176.4 or 192 kHz); combined with staggered clocks for the four S/H LPF circuits after the YM3029 and the PCM56P, that could yield two outputs running in stereo (16 bit) at 44.1 or 48 kHz.

Now I've never built something like this, but from a technical standpoint this is an interesting challenge.

I suppose the simplest solution would somehow use a pair of CS8406 digital audio transmitters to generate the two AES3 (or SPDIF) signals. Perhaps it might be possible to use the same sample-and-hold trick as in the SY99 to allow the use of only a single CS8406.

Alternatively, an FPGA (Spartan 3AN should be more than adequate) could be used to de-interleave the four CD-quality channels, which would then get sent to two CS8420 sample-rate converter and transceiver chips; these chips could take care of synchronizing the outputs with the provided wordclock (and convert the sample-rate, if needed) and they could drive the AES3 and SPDIF outputs. Additionally, the three-wire serial output could be sent to a Wavefront AL1401AG Lightpipe encoder, which could drive a multi-channel ADAT output. I think the Cirrus Logic datasheets refer to a Toshiba TOTX173 TOSlink transmitter, but I think the bandwidth on that is enough for SPDIF but not ADAT (which needs about double the bandwidth). Given that the TOTX147 is specced up to 15 mbps, it might be an option to use that for the SPDIF and ADAT outputs.

Anyway, I'm typing this from memory an my smartphone and I'm sure I'm making things to complex to be practical and to over-simplified to actually work 8)

TL;DR - the output resolution is only 16 bit (arguably 15) at 44.1 or 48 kHz. I really wonder if a digital output is really going to improve the audible output quality (when compared to a good external DAC).

If you just have some digital outputs that cannot sync to a provided wordclock (or if your interface cannot sync to that wordclock), you run the risk of audio drop-outs (crackling, etc). Something like the RME HD192 could solve that, but that thing costs like £1000/€1200/$1600 new!

After all this bla-bla, some fresh opinions would be welcome...


Gear list: Saleae Logic16, Rigol DS1052E (@100 MHz), some Xilinx FPGA boards, lots of screwdrivers and a decent soldering iron

JK1974
Member
Member
Posts: 85
Joined: Tue Oct 02, 2012 1:01 am

Re: Digital Output for SY77/SY99

Unread post by JK1974 » Wed Sep 11, 2013 10:18 pm

This sampling-frequency conversion might indeed be a problem - having been out of the 'audio scene' for more than 10 years, I hoped that this problem has been solved in the meantime by more sophisticated receiver/cheap high-quality converter chips.

I am interested in a digital output because I want the 'pure' SY99 sound. When I bought the SY99 20 years ago, I used a lot of sampling to get more 'modern' sounds. Like I wrote above, the sounds had more 'brilliance' when I played them back on the PC than on the SY99. Maybe it's the internal digital signal processing on samples, but maybe also the DA conversion. Furthermore, avoiding any analog noise is always of my interest as long as I use it as a studio instrument and not primarily for live performances.

Another aspect: Listening to lower frequency sounds, I often hear some background noise, also on AFM sounds. However, I can't tell from a technical point of view if this is also caused by the DA conversion or simply the drawback of those (earlier) digital synths that used a low sample playback rate (not only on waveforms, but also on the playback of AFM elements).



db7

Re: Digital Output for SY77/SY99

Unread post by db7 » Wed Sep 11, 2013 11:25 pm

The sampling rate of AFM coming out of the OPS3 is almost certainly 64 kHz. If we assume acreil on Gearslutz is right when he says self-feedback at high levels produces a superimposed tone at samplingRate / 4, my tests showed a tone appearing at exactly 16 kHz, suggesting a sampling rate of 64 kHz for the OPS3. And we know there’s a master clock at 6.144 MHz floating about, which is 64 kHz * 96.

Side note: that 96 does not indicate 96 operators per OPS3, for the 77/99 actually divide FM duties between two OPS3 chips, indicating 8 channels each (8 * 6 = 48 ops). Probably some other divisor is involved at some point, maybe even another clock. I posted a bunch of stuff about that and sampling rates over there, and some of it might come up again below, but I’m just going to offload it now anyway: http://www.gearslutz.com/board/9391049-post148.html I think what I say below is clearer…maybe. :P

The output of both OPS3 chips is funnelled into one of the two M3 chips, a.k.a. AWM engine (whether or not the user has tasked it with filtering the AFM). The sampling rate of the M3s is much less clear. I used to assume it was 48 kHz since that’s the maximal rate of waveforms users can load, but I think that was too simplistic. It almost certainly upsamples everything to some higher frequency to enable resampling, scaling of pitch, interpolation, and so on, not to mention that it presumably takes 64 kHz from the OPS3.

The other confounding factor is that, somewhere between the M3, the PANning processor, the DSPs, the MIXers, and the DAC, one or more sampling rate conversions might already occur.

However, I guess you guys are only interested in the final rate, of course, so my musings about earlier points are just academic.

Anyway, although we have yet to prove which, the final sampling rate is almost certainly either 64 kHz or 48 kHz. I think 44.1 kHz need not apply, being such an unrelated figure. We know the master clocks into the M3 and several other chips is 6.144 MHz, a common multiple of 64 and 48 kHz. Combining that with the details above, and assuming the DSPs (A) do not resample or (B) resample to a logically related frequency, the final signal is almost definitely one of these.

Determining which would require someone with a soundcard capable of recording at 96 kHz and a wide frequency sweep or suchlike. My hardware only goes up to 48 kHz, otherwise I would test this right now.
JK1974 wrote:Another aspect: Listening to lower frequency sounds, I often hear some background noise, also on AFM sounds. However, I can't tell from a technical point of view if this is also caused by the DA conversion or simply the drawback of those (earlier) digital synths that used a low sample playback rate (not only on waveforms, but also on the playback of AFM elements).
This is almost certainly due to bit- and sample-based (amplitude- and time-domain) quantisation of the waveforms stored in the chip. Playing at a relatively low pitch, you hear high inharmonic overtones introduced by the quantisation. These artefacts get progressively quieter as the fundamental rises and they get pushed further towards the inaudible end of the spectrum. Regardless, they remain in the waveform, and they are multiplied as soon as you modulate it even once with another quantised waveform.

It has been proven that the waves in earlier single-chip FM ICs are stored in 12 bits, just 1024 samples per cycle, and using logarithmic amplitude to boot. Having used such chips, I suspect I know what this does to the sound. Listen to almost any old 4-op bass, and hear that sort of sizzle added to the actual patch.

Having said that, in such old setups, the artefacts might be exacerbated by quantisation in the DAC; take the 10-bit mantissa (alongside a 3-bit exponent) in the YM3012, for instance. However, the DAC in the 77/99 is much better, of course. I bet it’s not a limiting factor in any way in that signal path.

Whether the waves in the OPS3 are stored at a higher resolution/rate than in the older ICs, and by how much, is still an open question. I suspect they are, but I’ve yet to analyse this against the older chips to get a better idea of whether/how the waves evolved. Even if better, they will still be imperfect, like any digital wave. So, either way, I strongly suspect the artefacts heard in AFM are due only to the quantised waveforms in the OPS3, not the DAC at all. Otherwise, you’d hear the same artefacts in AWM too, no?



JK1974
Member
Member
Posts: 85
Joined: Tue Oct 02, 2012 1:01 am

Re: Digital Output for SY77/SY99

Unread post by JK1974 » Fri Sep 13, 2013 1:25 pm

Determining which would require someone with a soundcard capable of recording at 96 kHz and a wide frequency sweep or suchlike. My hardware only goes up to 48 kHz, otherwise I would test this right now.
Would you generate this sweep with AFM or is your idea to upload a 48 kHz sample and pitch it to a higher frequency and then record it at 96 kHz?



db7

Re: Digital Output for SY77/SY99

Unread post by db7 » Fri Sep 13, 2013 2:14 pm

I was thinking of using AFM with a wide PEG. Something equivalent could probably be done with AWM, but the best way to check this is to use a sine wave, and since AFM already has one of those by default whereas AWM doesn’t, why reinvent the wheel?

Presumably the final sampling rate coming from the analogue outputs (and thus indicative of the sampling rate going into the DAC, which you hope to tap) is not higher than the 64 kHz coming out of the OPS3; again, I suspect it’s either the same 64 kHz or otherwise 48 kHz. So, the point at which the sweep disappears or begins to alias downwards would indicate whether the final output into/from the DAC is sampled at 48 kHz or 64 kHz (or, less likely, something else).



db7

Re: Digital Output for SY77/SY99

Unread post by db7 » Sat Sep 14, 2013 1:55 pm

Another big post. 8O To get it out of the way, I and I’m sure other people would appreciate if someone with hardware supporting recording at 96 kHz or greater can run a PEG sweep to determine where aliasing begins and hence predict (I think!) the final sampling rate of the 77/99. [edit] JK1974 provided that test on the next page; plus, I now have an interface that can record at 96 kHz. [/edit] That said, I couldn’t stop myself testing. In the second bit, I describe the method and why my tests can’t indicate the final sampling rate.


Self-feedback indicates the sampling rate of AFM/OPS3/EGM2 in the SY77/TG77/SY99 is 64 [edit] no, 48! :P [/edit] kHz

[edit] acreil investigated further and determined that the rule is frequency divided by 3, not 4, which indicates a sampling rate of 48 kHz for AFM in accordance with JK1974’s test on the next page. In light of this, my test here still proves the conclusion, just with a different initial proposition. ;) [/edit]

Earlier, I said this:
db7 wrote:The sampling rate of AFM coming out of the OPS3 is almost certainly 64 kHz. If we assume acreil on Gearslutz is right when he says self-feedback at high levels produces a superimposed tone at samplingRate / 4, my tests showed a tone appearing at exactly 16 kHz, suggesting a sampling rate of 64 kHz for the OPS3.
And here we have the evidence. First, the background again. In my next reply in that thread, I asked acreil what this really meant:
If this is definitely accurate/informative, what settings will produce the relevant frequency? Do you mean that, as increasing self-feedback turns a single op from a sine into an almost-sawtooth and then into something distorted, that distortion tends to collect around fs/4? Anyway, I’ll experiment and see if I figure it out, and I’ll post any implications I find.
Having gotten no reply, I forged ahead, and it turns out my assumption was correct. Here are sweeps from self-feedback levels on a single op with outputs of 127 down to 89 and notes E0, C3, A3, and G6. These show clearly how intermediate levels in this range produce a superimposed tone at exactly 16 kHz.
selfFeedbackToDetermineSamplingRateFrom127To89AtNotesE0C3A3G6.png
I can’t vouch for acreil’s reasoning, but assuming it has a technical basis, why would this occur? Why a tone at 16 kHz for a sampling rate of 64 kHz? If I had to guess, I’d say this effect is a form of self-oscillation with the 64 kHz divided by 2 twice by (A) the sampling theorem and (B) Tomisawa’s anti-hunting filter. The latter is used in all PM synths by Yamaha to stop self-feedback going crazy: the input from feedback is actually the mean of the previous two samples, not just the most recent sample.


Frequency response: no built-in EQ, and the reconstruction filter starts at a very reasonable/normal ~16 kHz

Back to the latest two posts by myself and JK1974, as well as this older post by him and subsequent ones with the same sentiment—
JK1974 wrote:Could it be that the YM3029 includes some kind of EQ to give the SY77/SY99 a "certain kind of sound"? 20 years ago, I had the feeling of a lack of "brillance", and so I often used a EQ to boost the higher frequencies on my own samples before I transferred them to the SY99.
—I got to testing, spurred on by having done something very similar very recently with my AN1x. As I’ve said, I can’t test my earlier question of whether the final sampling rate is 48 kHz or 64 kHz: my interface has a maximal rate of 48 kHz, so the fact that my results show imaging (reflection) down from 24 kHz says nothing about the rate of the 77/99: it’s equally likely merely to be imaging that escaped the recording hardware’s analogue-to-digital converter (ADC).

But I can test the frequency response of the analogue outputs and hence indicate the nature of the post-DAC reconstruction filter, as I did for the AN1x. I found that the –3 dB point, often called cutoff frequency, of this filter in the SY99 is just below 16 kHz. We can think of this in rough terms as the point where the attenuation of high frequencies begins to become audible, beyond which their roll-off matters.

The patch used was a 1AFM mono voice, initialised to default settings, set to Algorithm 1 (for maximal signal-to-noise ratio, removing the compensation for multiple carriers; see the final section), and given a Pitch EG from –64 to +63 across 8 octaves. I played the highest note, G6, recorded the output at 48000 kHz / 32-bit (really 24 bits coming from my hardware), normalised it to full scale (0 dB), and analysed it.

Here is the waveform graphed by linear amplitude:
sinswp1.png
Here’s logarithmic amplitude, showing my annotation to detect the –3 dB point:
sinswp2.png
And here’s how I read the frequency of said point. Note that, although we know the waveforms in the FM chip are imperfect, at least some of the artefacts visible in these graphs might be due solely to the high sensitivity and/or the windowing function I used, not the synth. I’d need to cross-check this with other settings. The important thing is the fundamental is clearly visible and a lot louder than all the rest.
sinswp3.png
Side note: to prove the PEG works logarithmically, as one would expect, have a spectrogram with a logarithmic y-axis, showing the perfectly linear ascent we get when viewing it on that scale:
sinswp4.png
We can see the –3 dB point of the reconstruction filter is around 15.7 kHz. This matches very neatly my findings from other synths by Yamaha: most of their LPFs are set up for almost exactly this –3 dB point.

JK1974, maybe, if you have very acute hearing at high frequencies, this might explain why you felt you heard reduced treble on samples after they were loaded into the SY99. Having said that, this cutoff is pretty high, and I wager most people could not detect its presence or absence in normal material (read: not just pure tones with very high frequencies). But if you can, congratulations!


AWM/M3 does not perform additional filtering of its own

Because the original question by JK1974 concerned samples loaded by him and processed through AWM, to check whether anything else might be at play within this engine (M3 chip), I did another test with a sample of white noise to determine whether this section performs additional filtering. Here is the original sample generated by Audacity, 4 seconds of white noise at an amplitude of 0.8, which uniformly fills the entire 24 kHz (maximal rate in AWM samples = 48 kHz divided by 2 due to the sampling theorem):
noisePr2.png
And here is what I got back after sending this sample to the SY99, playing it through an initialised 1AWM voice at full volume with no envelope and no filter, and then normalising it back to 0.8 in Audacity:
noisePst.png
Whereas the original sample was around –41 dB across the whole spectrum, the output of it played from the SY99 falls to –45 dB by around 15 kHz. Accounting for statistical noise, this indication of the –3 dB point accords almost exactly with my findings from AFM, indicating that the roll-off is due only to the reconstruction filter and hence that AWM/M3 or any other downstream part apply no filtering of their own.


Compensation for multiple carriers was not new to the 77/99

To follow up on something I said above, I seem to recall reading someone implying that the reduction of carriers’ output levels when more than one is present, which aims to help prevent clipping due to mixing, was a new feature of the 77/99, and an unwelcome one at that. This is false. The service manual for the original DX7 (and hence DX9) makes it clear this was present in Yamaha’s very first FM synth. So, the 77/99 did not add this. I’m not sure whether the 4-op chips do this; this would have to be tested.



Post Reply

Return to “Yamaha SY99 Forum”

Who is online

Users browsing this forum: No registered users and 1 guest