Roland D-50 noise issue with PCM sounds, SYNTHE sounds fine

[frisc0]

Hi, new here on EEVblog, found this forum while searching for a solution for my Roland D-50 synth and hoping to get some help getting this fixed. Not very technical but know the basics. I've posted the same issue earlier on the VintageSynth forum but no solution yet, now hoping to extend the help from this tech-crowd .

The issue, copied over from VintageSynth forum:

My Roland D-50 has an issue playing patches that use PCM waveforms, but plays all SYNTHE-only patches just fine. PCM waveforms sometimes play fine for a few times and then start cracking or stop playing completely. Sometimes they only produce a 'click' and after hitting some keys it's like something is 'popping' open and the PCM waveform is played correctly again, but always with a cracking sound.

Chorus and Reverb chips are okay, since the problem only arises with PCM waveform patches and the SYNTHE sounds also go through these chips afaik and sound fine. That's why I think the DAC and everything towards the outputs is fine as well. So it's probably something before the DAC.

Now, where to look?

What I did so far:
- replaced the battery
- reloaded the factory patches via MIDI
- ToneRAM reset via key 0 + Data Transfer
- performed the internal tests, multiple times, memory is (appears) ok
- unplugged and replugged all connectors
- checked mainboard for any visible issues to components
- recalibrated the MSB trimpot via procedure in service manual
- changed voltage on the mains transformer from 220V to 240V (230V in my country)
- changed all capacitors on both PSU and mainboard (...)

I have the schematics and tried to back trace down from the DAC. I also suspected the PCM ROMs, but my guess would be if they're faulty, they would never produce a waveform, and the 'popping open' in my mind would more suggest an analogue component, like caps? Note I did not replace the SMD caps, but did measure them (in circuit) and they seem to be ok (all have same resistance and charge/decharge similarly).

One thing I forgot to mention: it appears that Reverb presets 17-32 influence the cracking patches, some reverbs in this range even end up in a sort of feedback loop resulting in lots of digital noise. Could that be a RAM issue (either tone RAM or working RAM)? Is there any way I can test that without replacing all RAM ICs?

I did some further measurements on the mainboard, on the resistor arrays (in circuit). Most RA's are measuring values that make sense (15K-ish) but on RA10 (next to the CPU) 2 of the 8 resistors measure 10K instead of 15K ohms. RA10 is attached to the higher address lines (Ah). Afaik, these RA's are used as pull-up resistors to make sure to always get a clear 0 (ground) or 1 (5V). Could these 10K resistors (if it's a correct measurement, being in circuit) result in random ROM/RAM locations being addressed and read and as a result, the distorted PCM patches?

To further narrow down the issue, I've set up 3 basic patches with only 1 partial, PCM only (structure 6):

1. PCM waveform 1, Marimba
2. PCM waveform 32, Breath
3. PCM waveform 100, Loop24

Tone 1 always sounds clear. Tones 2 and 3 are mostly distorted, however!

When I switch off and on the D-50, the first 3 or 4 key presses for tones 2 and 3 sound clear and only after 4-6 key presses they start cracking and sometimes failing completely (only clicking/popping).

Interestingly, using tone 3 (which is looped), after switching off and on the D-50, if pressed and hold for a long time, the sound remains clear without distortion! Only after triggering this loop tone multiple times it starts cracking.

In some cases, switching from tones 2 or 3 back to 1, this tone also sounds distorted for the first 2-3 keypresses and then sounds clear again and remains clear.

My conclusion from this behaviour: PCM ROM IC2 is ok, it can deliver the correct data. Something in between is failing. I'm suspecting noisy addresslines in the higher area's of the memory map, therefore affecting only the waveforms from nr 32 and higher... does this make sense and if so, where to look?
Maybe the pull-up resistor arrays on the addresslines? I've measured these before and saw some lower resistance values (10K instead of 15K), but that would end up with a solid 5V as well, right?
Could also be the 0,1uF smd decoupling/bypass capacitors near the memory ICs that I haven't replaced yet.

I'd love to get this awesome synth going again, so all help is greatly appreciated!

[floobydust]

Without a schematic there's not much help we can offer. The synth might have a different path for PCM sound generation. Roland got cute with their designs to keep speed up and cost low.
I would heat/cool some of the usual culprits. Take a can of freeze spray and cool some memory IC's, or a heat gun and warm them to see what aggravates the problem.
PCM sounds might always use the reverb chain and bad RAM there would corrupt the reverb tails, which can be what you are hearing.

[frisc0]

Thanks for your reply floobydust, the schematics can be found here: http://www.synfo.nl/servicemanuals/Roland/D-50_SERVICE_NOTES.pdf

As far as I can tell all sound data comes from the Synthe IC and leads through the Reverb and Chorus circuits. If these circuits/ram are bad, it should affect the non-PCM patches as well I would guess? Btw I think these circuits are always engaged, can't switch them off in a patch only set the level to zero.

The symptoms (crackling starting after a few key presses) seem to point towards some analogue component 'saturating/overloading', rather than a digital domain component? Anyway, I'll check with your suggestion on cooling/heating to see how that affects the issue, thanks.

[floobydust]

You have to be a sleuth and try things to really confirm where the problem is.
Some PCM patches might set the reverb to different settings.
I would suspect a bad DRAMIC uPD41416C-12 16Kx4 (4416) if the problem is the reverb, in the digital domain. If they're in sockets you can swap positions to see if a problem moves. Or just heat them up a little and if bad it will get much worse.
During the release, the reverb data will be cycling down to all 0's but during sustain portion reverb is not involved other than loading.

To fix these I usually start by inspecting the solder joints, keyboards get banged around and it's common to have fractured joints at I/O jacks and buttons etc. Use a magnifier and bright light to carefully look around. You can tap or press the boards with a plastic pen (while holding a note) and see if a bad connection shows up.

If you think it's an analog problem, ESD hits at the output jacks does damage the muting transistors Q1-Q4 so you might work backwards from the output back. But the problem is in both channels? Does the volume control affect the problem, headphones the same?

[frisc0]

Thanks floobydust for these leads and thinking along, I already did some detective work on this board looking with a flashlight in detail at all components and soldering joints, do not see anything suspicious. Unfortunately, the DRAM ICs are not in sockets, so I'll try the heating you suggested.

The 3 test patches I've set up are build-up from an initial/empty patch, but I'll check what that does with reverb settings. Those reverb settings (as well as other patch data) is stored in the (battery-powered) Tone Ram IC. However, since SYNTHE patches that use reverb sound just fine, I don't think it's in this circuitry.

The interesting behaviour I see consistently is that after power on it takes 3 or 4 keypresses before test-patch 2 (breath) or 3 (loop) starts crackling. It's always clear the first 3 key presses. And patch 3 which is looped, also starts crackling after 3 or 4 keypresses, but being activated and hold for a long time the first few keypresses after power-on, the sound remains clear during the hold-period. So it definitely is related to how often the patch is triggered and not how long the sounds is produced.

I suspect that with every keypress, the PCM data for that waveform is being re-read from the PCM-ROM(s) and the distortion starts only after 3 or 4 'reads'. Since this behaviour can be reproduced consistently, I suspect 'something' is overloading during this process, but I can't pinpoint it to a component... should I replace the SMD decoupling capacitors next to the PCM ROM ICs?

I don't think it's in the output circuitry, since the SYNTHE patches sound fine and go through the exact same route. Also ,there is no difference in the upper and lower channels or headphone vs output jacks.

I did see some forum posts on muting transistors being faulty, but that would also affect the SYNTHE patches, right?

[Audiorepair]

When you play a sound, does it read the PCM data from the ROM, or does loading a patch read once from the ROMs and arrange everything and play back from RAM?

Try looking at the CS and OE pins on the ROM's, (pins 22 & 24 of IC's 29 & 30), this may indicate one or the other, which may narrow things down somewhat.

[frisc0]

Good question, I see that the SYNTHE IC (DSP) does all the (digital) audio signal generation and the CPU appears not to have any part in this. The ToneRAM and Working RAM are attached to the CPU, don't think the SYNTHE addresses this same memory space. I did wonder whether delivering PCM data from ROM for processing could be fast enough, and instead you'd need RAM to achieve correct speed, but looking at the schematics, it looks like only the reverb IC is using the 41416 DRAMs? PCM sample's are small, like 2-5Kb, maybe the Synthe DSP has some internal memory? The D-50 is mono-timbral (okay dual, when using upper and lower parts of the patch and a keyboard split) so it only has to hold 2 waveforms max.

Off-topic, but just wondering: how does the CPU on the D-50 mainboard address all the memory (ROM/RAM) that it's using? I did some line tracing on the schematics and found the following.

- The CPU is capable of addressing 64kB (16 address lines) external memory
- The Program ROM is 64kB (not sure if everything is used, but all 16 address lines are wired)
- There is 32kB of Tone RAM holding the config of all 64 patches (essentially what you sysex bulk-dump), with battery-backup
- And there's 8kB of Working RAM for the CPU (for managing state/settings/MIDI buffer/etc?)

There is also 8kB of Chorus RAM and 48kB of Reverb RAM, but that doesn't seem to be directly addressed by the CPU, but by the dedicated Chorus and Reverb ICs. Just like the PCM ROM ICs A and B (2x128kB=256Kb, 18 address lines wired) are directly addressed by the Synthe DSP.

Question is: since there is already 64kB of ROM, how can the CPU also address the 32+8=40kB of RAM? My guess would be at power-on the CPU reads the Program ROM and initializes the other devices like Synthe DSP, Chorus, Reverb and Dynascanboard and then magically remaps memory so that part of the 64kB map points to this 40kB of RAM. I see some lines of I/O port 0 going to the chip enable lines of these RAM ICs, that might explain it. Anyone with more info on how this works exactly?

And another question that popped up while reading the schematics: the digitial audio data is 16bit throughout the D-50 (Synthe > Chorus > Reverb > DAC). However, the service manual speaks of 20-bit D/A conversion. The PCM54 IC is 16bit, and there is an additional 4bit R/2R circuit that is attached to the lower 4 bits of the 16bits audio data, that is again fed into the PCM54 as a feedback signal??? What's the purpose of this, how does it affect the output sound?

[Audiorepair]

The Synthe chip IS connected to the 41464 DRAMS by DR8 - DR20, so it may be capable of reading from the ROMS and writing this data to the RAMs.

The majority of these lines are connected to IC's 1, 2 and 3, so perhaps if one of these is faulty, it might affect only the stored PCM data.

[frisc0]

They are connected, but aren't those lines holding the digital audio data? I see the Write Enable lines of the DRAMs attached to the reverb IC. I'll check what heating them up does to the distortion (don't have freeze-spray, yet) and also the ROMs.

[Audiorepair]

Think you're right about the Write Enable pins, clearly you've spent much more time with the schematic than I have.

So it looks like the PCM data is always read directly from the ROMs then, at every key press.

Does that not suggest it is this process that is failing?

[frisc0]

My knowledge on hardware / electrical components is very limited, could a ROM IC be faulty and behave like this, at power on producing the correct data and at some point failing? Is there anything to be done to fix that other than replacing the IC? What other components surrounding this circuitry could result in this behaviour? Only components I see are the pull-up resistor arrays and the decoupling caps, could other parts be involved?

[frisc0]

Ok, so I've heated up the (ROM/RAM/DAC/DSP) ICs one by one while checking the sound of the faulty/distorted PCM waveforms. Unfortunately, no apparent effect. Still the 'lower half' of the PCM waveforms sound clear, while the upper half sounds distorted. It MUST be something with the PCM ROMs, right? I'm thinking about changing the two decoupling SMD caps (0.1uF) next to the PCM ROM ICs one by one. However I do not have substitute SMD caps, but I do have regular ceramic ones. Are they interchangeable, can I use these?

[Audiorepair]

Yes.

SMD devices are just equivalents designed to be small, cheap, and easily placed on a circuit board during manufacture.

Your ceramic caps might well outperform them, but this isn't really a critical issue, you just might need a bit of decoupling to be sure this is not the problem.

[frisc0]

I've changed the SMD decoupling caps near PCM ROM A and B to regular ceramic ones, but unfortunately, the issue remains...

Anyone got a suggestion where to look next? Running out of ideas... 

[floobydust]

Does it malfunction with power on time (heat) or just the number of times you play a sound?
What do you mean by "lower half" and "upper half" of the PCM waveforms?

If the least-significant bits sound fuzzy, distorted or warbling, beeping chirping (for all sounds), then I have seen the DAC IC need calibration.
You'll hear that as the sound fades off on the decay portion, before the muting engages. The quieter the sound is becoming, the more digital noise is heard up to the point the muting turns on.

I would step back- check the power supply voltages are exact and not drifting, and no memory IC is much hotter than others. They can draw more current.
If it's a memory IC problem, get some cold spray and try that. Trying to figure out exactly how the synth works just leads to a goose chase.
Seems to be a memory problem for the reverb but maybe it's something else.
I would read out the ROM's to archive and test the contents. It's not a capacitor problem yet.

[YetAnotherTechie]

You can do your hot test also on the glue logic, don't limit yourself to the big chips.
You said replaced the battery, had it leaked? it may have sweated, and got electrolite in your board, check the components around it, and the vias too, the acid can leak inside components and damage them, yes, it can go inside chips and resistors, resistor networks, etc.

[frisc0]

Thanks for your replies! Really appreciate the help.

Does it malfunction with power on time (heat) or just the number of times you play a sound?

Not related to heat or power-on time, it's malfunctioning (distorted sound) from the moment the device is powered on and after 3 or 4 key presses, but only for patches with waveforms #32 and higher.

What do you mean by "lower half" and "upper half" of the PCM waveforms?

Yes, sorry, that's my perception of how this works. There are around 64 waveforms available in the D-50 (and some loops using these same waveforms, counting up to 100 selectable waveforms). The first set of waveforms (lower-half, #1-#31) sound fine, no distortion. Waveforms #32-#64, upper-half, sound distorted. There are 2 PCM ROMs, A and B. That's why I thought this could be related, pointing to one PCM ROM IC. However, from power-on the 32-64 waveforms sound clear and after 3-4 keypresses sound distorted, so the ROM can deliver the right data...

If the least-significant bits sound fuzzy, distorted or warbling, beeping chirping (for all sounds), then I have seen the DAC IC need calibration. You'll hear that as the sound fades off on the decay portion, before the muting engages. The quieter the sound is becoming, the more digital noise is heard up to the point the muting turns on.

Checked this using the test/calibration process explained in the service manual, there is a MSB adjust pot. No distortion at DAC level, it's generated earlier in the process.

I would step back- check the power supply voltages are exact and not drifting, and no memory IC is much hotter than others. They can draw more current. If it's a memory IC problem, get some cold spray and try that. Trying to figure out exactly how the synth works just leads to a goose chase.

I'll check the voltages again, also on the ICs themselves (Vcc). And you are right about the goose chase however I really enjoyed trying to break down the schematics and understand how the device operates. In fact it's not much more than an 8-bit eighties PC with a soundcard with dedicated DSP...

I'd hoped understanding the workings would also help in tracing down the issue, since it appears to be originating in the digital area, not the analogue/audio/amp circuitry.

Seems to be a memory problem for the reverb but maybe it's something else.
I would read out the ROM's to archive and test the contents. It's not a capacitor problem yet.

How would one read out the ROM's? They're soldered on the mainboard unfortunately, not in a socket. But even if they were placed in a socket I would not know how to do this..?

[frisc0]

Thanks for thinking along, appreciated!

You can do your hot test also on the glue logic, don't limit yourself to the big chips.

What do you mean by the 'glue logic'? Small SMD ICs? Or also the discrete components?

You said replaced the battery, had it leaked? it may have sweated, and got electrolite in your board, check the components around it, and the vias too, the acid can leak inside components and damage them, yes, it can go inside chips and resistors, resistor networks, etc.

As far as I can tell the battery has not leaked, the battery area is clean. But I'll inspect this again a little closer, thanks for the suggestion!

[frisc0]

So I've measured voltages from the PSU again, nice clean +11.48V and -12.37V and a steady +5.02V. Measured on the IC's the +5V is also good. Checked the battery area again, nothing suspicious. Also checked heat of the ICs after 30mins of playing a looped note, but nothing remarkable.

I think it's the Synthe IC that's failing, specifically, some buffer memory on this chip. Let me explain why.

Not sure whether it's the test patches I've created again from scratch, or if it's the change of (SMD to regular) caps, but the behaviour of the distorted/clear patches seems a bit different now.

I've setup 3 PCM only (struct 6) patches:
1. upper only, PCM waveform #1 (marimba)
2. upper + lower, PCM waveform #1 and #2
3. upper + lower, PCM waveform #32 and #33

A) From power-on, playing 1. or 2. sound clear, no matter how many notes are played.
B) From power-on, playing 3. sounds distorted right away and stays this way with consecutive notes.
C) Playing 3. one note and then 1. or 2. these sound clear right away.

D) However, after having played 3. for a few (distorted) notes, playing 1. or 2. again, the first few notes are distorted, but then sound clear again...

E) Strange thing is, when having played 3. for a few (distorted) notes, then playing 1. one note (distorted), then playing 2. one note (distorted), and repeating this between .1 and .2 the distortion remains UNTIL either .1 or .2 is played a few times in a row after which they sound clear again...

My conclusions from this:
- the distortion can also occur on PCM waveforms 1-31, but only temporarily
- the distortion on PCM waveforms 32 and up is immediate and consistent
- the PCM ROMs are not the cause of the distortion, why then would the distortion of .1 and .2 appear and disappear again (test D and E)
- the Synthe chip must have some on-chip working memory or buffer that is failing partially

Does this make any sense?

I don't think the 6 DRAMs are involved in the issue. These are only attached to the reverb IC and reverb works fine (and is independent of selection of specific PCM waveforms).

Other thought that crossed my mind: could it be noise data or address lines between ROM and Synth? You'd probably need a scope for that, right? However, the consistency and reproducibility of the issue does not point in the direction of interference of some kind...

Do you think it's indeed the memory of the Synthe IC that is failing (then I'm done, since it's not repairable) or could it still be something else? Looking forward to your views/ideas, thanks!

[floobydust]

What is unique about waveforms #32-64 after three keypresses is anyone's guess. It's distorting strictly the raw (source) PCM sounds? Not the programmed patch which is the sum of many settings?

I believe this synth generates analog sounds augmented with PCM samples to add realism (during attack/decay), as well as adding reverb and chorus.
So in real time it has to read/add the synth+PCM+reverb+chorus data and output to the D/A.
I'd suspect the scratchpad memory where this is done, and based on common failures in audio gear, it's bad DRAMs. They look like they're used as shared memory between the multi-processors.
Or, it's the memory with the voice assignments that is failing. The program for a particular patch might be getting corrupt and pointing to the wrong PCM data or reverb settings etc. after three reads. As if a cartridge is bad. The cartridges (and port) get hit hard with ESD and there is no protection there.
To check a memory addressing/chip select problem, you would need an oscilloscope at least.
12 memory IC's:
Reverb DRAM IC1-6 uPD41416C
HM6264 chorus SRAM IC7
HM6264 working SRAM IC23
HM62256 Tone RAM SRAM IC24
LC3517AS SRAM
PCM ROM A IC30, ROM B IC29, Toshiba TC532000P 2Mbit custom mask ROM.
Program ROM, Firmware IC22 MB27C512 EPROM, rev unknown.

If you think it is the source PCM sounds going bad, then it might the ROM's or their addressing that is failing.
I think people use a TL866 chip programmer to read out their contents, and then reprogram new IC to put back in place.

I stumbled onto this which is weird, MAME and the D-50?
https://github.com/mamedev/mame/blob/master/src/mame/drivers/roland_d50.cpp

Cartridge midi dumps

[frisc0]

What is unique about waveforms #32-64 after three keypresses is anyone's guess. It's distorting strictly the raw (source) PCM sounds? Not the programmed patch which is the sum of many settings?

Both the raw PCM sounds as well as the programmed patches that use one or more of these nr 32-64 waveforms sound distorted.

I believe this synth generates analog sounds augmented with PCM samples to add realism (during attack/decay), as well as adding reverb and chorus.
So in real time it has to read/add the synth+PCM+reverb+chorus data and output to the D/A.
I'd suspect the scratchpad memory where this is done, and based on common failures in audio gear, it's bad DRAMs. They look like they're used as shared memory between the multi-processors.

I was thinking the same thing, with that scratchpad memory or temp buffer to do the calculations on and provide the digital audio stream. According to the block diagram in the service manual there is no memory address path between the Synthe IC and the Reverb DRAMs, only the digital audio stream goes to these DRAMs and there is no difference between the origin of the digital audio stream (PCM ROM A or B or SYNTHE waves). However, the Synthe IC is attached to the main address bus (high/low, 9 lines in total) and the data bus. I did not realize earlier, but the Working RAM or Tone RAM could also be used as this buffer/scratchpad memory and could be faulty...

Or, it's the memory with the voice assignments that is failing. The program for a particular patch might be getting corrupt and pointing to the wrong PCM data or reverb settings etc. after three reads. As if a cartridge is bad. The cartridges (and port) get hit hard with ESD and there is no protection there.

Good point, so it could also be the Tone RAM holding the patch settings. I did heat up the RAM ICs without any noticable effect in the distortion. I also saved the test patches to different bank/patch locations with no effect. Is there any other way to find out if a RAM IC is working correctly or is failing? The D-50's self-test checks ok, but not sure how thorough this memory test is...

If you think it is the source PCM sounds going bad, then it might the ROM's or their addressing that is failing. I think people use a TL866 chip programmer to read out their contents, and then reprogram new IC to put back in place.

I guess it's still possible that one of the PCM ROM ICs is failing, but since waveforms 1-31 also (temporarily) get affected by the distortion, I think the problem is somewhere else, like a memory buffer. Still unclear whether the Synthe IC has internal memory of it's own.

I stumbled onto this which is weird, MAME and the D-50?
https://github.com/mamedev/mame/blob/master/src/mame/drivers/roland_d50.cpp

Cool... the D-50 as a classic arcade?
Or the start of building an emulator?

It does shed some light on my earlier memory mapping questions, not sure if it's correct.

void roland_d50_state::d50_mem_map(address_map &map)
{
   // Internal ROM is enabled at 0000–1FFF (+5V pullup on EA pin)
   map(0x2000, 0x7fff).rom().region("progrom", 0x2000);
   map(0x8000, 0xbfff).m(m_eram, FUNC(address_map_bank_device::amap8));
   map(0xc000, 0xdfff).ram();
   map(0xf400, 0xf7ff).rw("keyscan", FUNC(mb63h149_device::read), FUNC(mb63h149_device::write));
}

Btw I see now that the PCM ROMs are 512kB in total, not 256kB. There are only 18 address lines on the ROMs, but there are 2 extra lines connected to the CS of the ROMs (duh...).

[cheater]

Try asking about this on the synth-diy mailing list - there's a few folks out there who repair digital synths.

[floobydust]

I would get some freeze spray and see if that changes anything, otherwise you'll need a scope to look at waveforms.

[frisc0]

You can do your hot test also on the glue logic, don't limit yourself to the big chips.

https://en.wikipedia.org/wiki/Glue_logic

So, that's the simple TTL logic ic's... Every day you learn something new   I'll try that, thanks.

[frisc0]

Funny, I've found a thread describing exactly the same issue as my D-50 has in this German forum:

https://www.sequencer.de/synthesizer/threads/roland-d-50-verzerrter-klang-aber-nur-bei-ein-paar-wenigen-patches.146274/

TS notes that 'Bei den Factory Patches passiert dieser Fehler nur bei den Patches: 22- Bass Marimba, 42- Vibraphone, 48- Piano Fifty, und Patch 62 -Oriental Bells.....'. Looking at what PCM waveforms are used for his failing patches, they all are in the 1-31 range, so possibly, the other PCM ROM IC. My D-50 is failing for waveforms 32-64. Could point to faulty PCM ROM ICs since there are two, A + B?

Unfortunately, no solution in that German thread, but it ends with the same advice: 'use freeze-spray to localize the failing component'... I'll get me a can...

Should it indeed be the PCM ROM ICs, these are not easily found. Could an EPROM with the right contents burned act as a replacement? Of course one needs a dump of these ROMs as well... Which can easily be found here https://dbwbp.com/index.php/9-misc/37-synth-eprom-dumps, specifically here: https://dbwbp.com/synthbin/Roland%20D50%20PCM%20ROMs.zip!

I'm confused: the PCM ROM IC dumps are 512kB each, but isn't 2Mbit 8-bit 256kB? Why would these dumps be 512kB per ROM?