Power rail noise in cd player

[RetroAudio]

I'm working on a cd player and have been getting some scope traces to see what noise levels exist at various points and have finally decoded (somewhat anyway) what I'm seeing. I'm using a 300mhz analog tek scope and get traces of the clock frequency (16.9344mhz) which occurs at probably what would be normal levels. Limiting the BW to 20mhz is what I use for this measurement and there does not seem to be much else on the trace. However, using the entire BW, I'm picking up all kinds of freaky stuff. I'm sure partly what I'm getting is just noise pickup by the probe, but I'm pretty sure there's a real signal in there after having messed around with it enough. Probably the most interesting reading came directly at the 5V regulator output in the form of a high frequency decaying toneburst waveform. Weird, huh? I counted the period and it measures about 250mhz. Other probe points showed a more steady frequency and resembled just noise until I played around with the probes enough to get a more clear reading of a steady frequency going on in there. I'm just curious if there is indeed some signal content in the digital portion of the circuit that warrants this measurement, or if there's some spurious resonance thing going on here, or just my measuring is introducing problems where none exist. Anyone else have similar experiences? I did some minor playing around like using a ferrite bead I had on hand in place of one of the jumpers being used on the power bus (I make no claims that I'm working on the best laid out circuit) but it actually seemed to make one test point a little worse which I thought was kind of odd. I'd really like to tame these power rails, at least for the sake of getting clean power to the clock pins.

I did some review over "how-to" measurements for digital circuits and this includes keeping the ground lead of the scope probe close to, or even partly wrapped around, the probe so as to minimize noise pickup. For example, taking a reading of a power pin could be done by connecting to the decoupling cap directly at it's + and - leads, instead of placing the 2 leads inches apart (which is otherwise more realistic but not good practice). I did notice differences in noise levels by using various methods. It is hard to get it the way you want to since layout doesn't always cooperate.

Anyone?

[RetroAudio]

Well, I have been busy tracing the circuit, finding out more, and installing new oscons (UCC PSA) and ceramic bypasses on the digital board trying to tame the rails and have succeeded to a degree. I was able to improve noise levels by a number of magnitudes. My #1 aim was to get clean power to the clock, but I felt it was important to get it anywhere I could once I got in and started testing. A separate clock p.s. that seems to be so popular isn't really possible in my cd player since it's traces are combined with others, namely the input serial data (w/clock) as read from the disk. Anyway, there does seem to exist a noise floor in circuit since even ground has a level. I'm kinda baffled by it so far and am still dealing with it I guess, even after having made some custom copper tape cabling for a makeshift ground plane.

I used this as a stopping point to listen and I thought I could detect a rather apparent change right off even before the system got warmed up. What I thought I heard was foremost a change in presence, or realism to the instruments. It seemed to become even more pronouced as the system warmed up. I'd have to listen more to make more of an evaluation, but the major change I think I have picked up on. I know that getting cleaner power to the clock can only help and perhaps this is mainly what I detected, but I also suspect that cleaning up the 1's and 0's might have made a difference, for example in altering the inherent hoise content. This may be an outrageous comment. I don't know cause I'm still learning what goes on inside a cd player. The only thing I do know is that p.s. rails have been improved by getting the noise levels down. I see no low frequency modulation of any sort or any additional high frequency funny business going on looking back and forth between idle and playing conditions just with the methods of using a wide BW scope, so perhaps all remains well. It certainly sounds like it anyway. And I havent even started with the analog board.

[Sonusthree]

Hi Retro,
It's probably not important but I just wanted to say that I am very interested in your results. I'm just starting to tackle power supply noise within my player but I don't yet have a proper 'scope. I guess I'm flying blind!!
I'm quite surprised that no-one else has chimed in yet. Keep up the good work and I'll be following your progress very closely.

Kind regards,
Martin.

[RetroAudio]

Hi Sonus - Thanks for a reply. Doing the digital board has been interesting and a change from doing analog. It's a different world for sure and the wide BW scope has been a big help in seeing what I have. When I got it a few years back it was mainly for seeing oscillatory behavior in analog. I think it's a must have for any serious work in digital - just my opinion however. The caps I used turned out to be about the same in value as the stock caps, it's just that oscons are much better. The value must have something to do with operation too since I installed a number of different values to see what worked best. I thought that lesser value ones of better quality would be the best but that does not seem to be true, at least for this particular player. I guess one guide to be used is that we should remember that we're dealing with MHZ here and that 3-terminal regs do not really operate in this region effectively, so it's all about the cap taking over and doing the job. I haven't decided at this point if I'm going to pursue the digital board some more or go on to the analog. I did see something interesting during my last testing. Those decaying tonebursts I mentioned in the first post look like they might be a product of a clock (not the 16mhz one) I'm picking up from the data receive chip . It's a 5mhz clock, so the board label says, and it looks like it's been manipulated into a toneburst shaped waveform, or just the envelope rather, without the fill. I was expecting the usual sinewave shape for a clock signal. It's on it's own 5v reg. too, so perhaps I'm picking up ground bounce (or whatever) at the other 5v reg. output. I'm sure I'd uncover more if I went looking, but it's not easy taking it all apart to make access for something that is not meant to have access, only to turn around having to do it all over again. I'll wear out the connectors if not carefull. More later.

[RetroAudio]

Oh yeah, something else I forgot to mention - I wondered about the XO (crystal oscillator) and how they are supposed to be so super sensitive to voltage supplies and vibration. Well, for a vibration test, I lightly thumped on the crystal itself while monitoring the 16mhz clock, and guess what - I didn't see any difference. I was expecting to see that fast of a signal change and become blurry, change levels, or something. I didn't and it stayed rock steady. As far as being a pure sign wave, it was not, but was a distorted one, although better looking than some clock signals I have seen published. This clock signal travels through a
wire to get get to the DAC clock input and is terminated with a 30pf NPO ceramic, sort of like traveling through an LC filter. I did not have access to see what the cleaned up clock looked like and will have to make access for that when I start working on the analog board. Just from this little test, I'm kind of wondering if vibration control isn't more for the disk drive than anything else. My unit does have an iron plate mounted below the drive bringing the weight up to about 15 lbs, a little heavier than your typical player weighing in at 10lbs, so the manufacturer must have thought it was good for something. Another test I thought of after the fact was to take a photo or something of the clock signal before the mods just for a simple comparison. Not sure if it's possible to detect a real difference like this or not.

[martin clark]

take a photo or something of the clock signal before the mods just for a simple comparison. Not sure if it's possible to detect a real difference like this or not.

It's not - the variation that matters is in the cycle-to-cycle timing, which is several orders of magnitude below the jitter in the o'scope display timebase. But don't let that stop you trying!

I usually limit the scope to 20Mhz or rather less when looking at PSU noise, to be sure I'm really seeing what is going on rather than just my sloppy technique. I also like using a simple x10 opamp-based preamp between the probe and scope, with an input filter limiting bandwidth to 50-100Khz. An OP37 and 9v battery is a cheap and effective start. It helps on two counts - with a suitable inamp you can see well below 100uV/div, and - esp. for the analogue stages - the bandwidth filtering gives very clean traces, which helps pin-down which rails need better regulation.

[RetroAudio]

Martin - Thanks for speaking up. I remember reading something on jitter and that some specialized gear is needed to evaluate. And yes, this 300mhz BW has been interesting to play with, not knowing if it was a curse or benefit at times...... I think it has revealed some things, but I am forced to insert a "fudge" factor here concerning readings. Collectively, they have made for some interesting comparisons that can perhaps serve as a guide at least. I like the idea of adding on a boost preamp to read lower levels when the time comes. I was able to get noise levels down to the order of 10-20mv on the clock noise, several times lower than what it was stock. As for the high freq. noise, I was getting 50mv - 100mv levels, also much improvement over stock. However, I'm not sure about absolute levels given the noise I pick up just on ground itself and probe pickup. Doing something like you suggest looks good to me, just have never had the need to go this far before.

One more thing about voltage regulators: I got into noise figures when I chose new ones to install (what the heck, I was just having fun and concerned about stability with the new caps, not being sure the stock regs could handle it). Anyway, at the levels here, the micro-volt noise level ratings appear to be swamped out by the p.s. noise levels. But another way of thinking is that perhaps it could still make a difference as to the total noise it does inject into the signal. I mean, why add more if you don't have to? Every bit counts and after all, the purpose of this exercise is to decrease noise any way possible. The new regs are staying in. I'll probably do some more thinking on this and come back to the digital board again. I just wish access and making measurements was easier on some of the signals.

But, overall I'm just trying to have fun and not get my knickers riding too high. I'm pleased with the results so far.

[martin clark]

Anyway, at the levels here, the micro-volt noise level ratings appear to be swamped out by the p.s. noise levels. But another way of thinking is that perhaps it could still make a difference as to the total noise it does inject into the signal. I mean, why add more if you don't have to?

Completely agree, with the caveat that I'd take a reg with lower output impedance but more noise over a very quiet, higher-Z reg every time. If the noise is random (ie not correlated with the load) it doesn't seem to matter much...

Remember that the reg's noise rating is the 'self noise' measured on the output of the reg with NO LOAD. That's where the difference comes from - the far-greater noise on the rails is mostly due to the noise contributed by the load(s) and the ways in which this noise is NOT absorbed by the regs (limited) bandwidth, output impedance, the decoupling caps, poor PS routing/layout (long traces) and so on and so on...

In fact, even if you had a perfect regulator with zero output impedance over an infinite bandwidth (!) I think you'd still see some noise due to the last two items in particular...

[Extreme_Boky]

It is very refreshing to see some people still tackle the real problem in digital audio (well, in any audio) – noise.
I did the similar thing long, long time ago and together with reading various info on this topic, learned a fair bit about mods and how to get the most out of the spent dough. I experimented a lot in the exact same way as you are doing now. I found, for example, that uPC generate most noise and that noise had to be prevented from entering the common ground plane at all costs.

The real question is: what’s the point in spending a lot of money on expensive decoupling capacitors when all they do is couple the power supply rail noise to already very noisy ground plane? There are some good ground-plane designs out there, but only a few – MF does very good job in this regards.

As you mentioned already, reducing the noise by the order of magnitude (especially around digital receivers, decoders, upsemplers, DAC’s, clock signal generators…) brings tighter, better focused and extended sound, which is also better positioned in space around speakers. Re-designing the ground plane (I use very wide pure silver ribbons for this purpose) and careful observation of noise around digital IC’s (and proper selection / positioning of decoupling caps) can lead to fantastic results. After a while, you’ll know exactly which capacitors to use, and how. But before that, the real trick is reducing the ground plane noise. You’ll hear different opinions and read different solutions, and this is all good. However, your measurement (and your ears) will be in contradiction with vast (common?) thinking of many so-called (self-proclaimed) experts. Try as many options and apply as many approaches, but stick to what your measurements are saying to be the best, and follow by the listening test. Another tip: if you are tired, the low jitter, low noise results could sound too clinical; if you are well rested – they’ll sound amassing. A lot depends on the rest of the equipment. You can fine tune and merge the components of the system together by properly selecting the power, interconnects and speaker cables.

All sad, I decided long time ago to use the least amount of digital circuitry possible to obtain digital to analog conversion – receiver and DAC.

Boky

[RetroAudio]

Crowbar - I looked into using shunt regulators and zeners starting out but put them on hold since there was enough going on. I also didn't want to deal with the extra current needed to operate them. And you're right about shunts and the mhz dealt with, so caps are the issue here (for now).

Extreme - nice to hear from someone with similiar experience. The approach I'm taking as an EE is that of what one would read in one of the trade publications like EDN and app notes which are full of great info. But I'm probably less hardcore than most as I do favor a fun approach at all times, so I take the freedom to play and mull things over. I'd also like to think I'm in tune with the caps and other things that are suitable for audio. So far I've just been using oscon-like caps and NPO/COG ceramics. I have done some more testing and listening since my last post from the work I've been doing on this "data receive" chip with a 5MHZ clock. This chip is responsible for getting the data from the drive and clocking it to the 16MHZ main chip. I upgraded the caps on this 5v bus and put in some bypasses. One thing I noticed is that a 330pf ceramic was on this p.s. rail already and I wondered why. I thought it was a rather odd value to have at first for a clock of only 5MHZ, but I also knew more was going on and I'd have to pursue to find out instead of leaving it at just a guess. After more probing I found out the chip was processing this 5mhz into what appears to be a gated toneburst, and using only the first complete cycle (as spikey as it is) as the clock to the 16MHZ chip. Power rail measurements on this 5MHZ chip show a noise with spikes at intervals. With the new caps, seeing a reduction in noise measurements using the entire BW was inconclusive that I could see, but using a 20MHZ BW showed a 25% reduction from 40mv to 30mv. It was here that it occured to me that bypassing with the PROPER value of cap was paramount and that what it is I'm dealing with is, put simply, a tuned circuit. I have been using an approach in bypassing so far that consists of a .1uf and .01uf NPO cap soldered onto the oscons. I'm sure this has been helping to a degree since my measurements seem to indicate it, but knowing the frequencies involved at every juncture and bypassing with the corresponding cap is what is really needed, and that my method is not fully correct but only a rough approach. The 330pf cap as noted on the p.s. rail would seem to be a bypass meant for the outgoing clock signal, which is an intermittant but rather high frequency indeed. For keeping things cheap, I think the manufacturer knew full well what they were doing and I have been trying to pay attention to their intent all along, but it has been difficult at times. Sometimes I don't know if they're just being cheap or using their smarts.

Now, one more observation is that my previous finding of a toneburst at the 5v reg's output for the 16MHZ chip doesn't look so nuts afterall. It looks to be feeding the power requirements for it's input data and clocking setup. Big surprise, huh? Well, it was confusing at first but since I know more now, it makes a bit more sense. Less muddy at any rate. I'm also of an idea that perhaps a smallish picofarad cap at strategic points of this power rail may help. I might start at the 5v reg's output and give it a check to see how the trace is effected. I'll also have another look at the data stream and clocks to get more precise measurements and values and go from there. Ideally, a spectrum analysis of the p.s. noise is what's needed which would give way to proper bypassing, but I don't have that luxury. I can makes some guesses though and go from there. It looks like I might be covering the same territory as what Extreme has been through, but it's common fare for digital circuit work, so nothing really new. One other thing I noticed is that the value of caps at one location do indeed effect noise measurements at other locations. Case in point: On the 16MHZ chip, I decreased the value of capacitance on the analog power pins that occupied on side of the chip only because larger ones that emulated the stock values would not fit. Well, I moved some of the interfering parts on the other side of the board and installed stock-valued caps and noticed a decrease in noise on the adjacent and opposite side of the chip, namely the clock p.s. and various outputs. That was a good move since reducing the clock p.s. noise was the object here. Point is, on a daisy chained p.s. rail, caps feed the next one in line and interact with one another.

So, how did this additional "upgrading" sound? Well, on the first try my sinuses had been playing havoc and I was on some meds. The whole sound was absolutely horrible but it was me not doing so well. I listened the next day and things were much improved, but by how much is difficult for me to say. I didn't hear as dramatic an improvement as I did the first go 'round, but I do believe there was a little bit. If I had to describe, it'd be just more of the same as what I first heard, if anything. Certainly not worse. Nuances are standing out as much as ever.

I'll try to be back later with more if I can find anything noteworthy to do and see further improvement. Some of the traces are starting to pull up from desoldering, so my time is probably limited with this part. I do wish I could tune this to perfection though. I'd like to hear what one would sound like. Seems to me that could be the next step in player design and marketing, even if they did use cheaper parts. Nah, too logical.......gotta leave freaks like us wanting more....