Since things have been kind of slow around here for the past few days, thought I would report in on my experience with an IanCanada 45MHz audio clock module. It in currently installed in the prototype clock board, and its been running continuously for the past 4-5 days. Initial jittery sound has faded away as is typical for new clocks. A pic of the clock in the prototype clock board below, in case anyone is interested in how it looks (its the metal can with the green LED inside):
Clock board is driving an Andrea Mori FIFO board which then drives the dual Andrea DSD dacs.
So far the IanCanada clock has been A/B compared to the Accusilicon 45MHz clock using the same clock board. For the purpose of driving the Andrea FIFO board, the clock frequencies are divided in half using NC7SZ74K8X flip flops. The division is expected to improve phase noise by about 6dB. (45/49MHz clocks are used in order to provide the option of buffered outputs for external clocking of I2SoverUSB.)
The clock board 22/24MHz outputs were also compared with AckoLabs 22/24MHz clocks for driving the FIFO board.
Basically at this point the Ian clock sounds significantly less "blurry" to me than the 45MHz Accusilicon clock. Definitely a step up in that regard.
The SOA AckoLabs clocks were chosen for comparison mainly because they are small and simple to connect/disconnect and move out of the way as needed. AckoLabs measured phase noise is similar to Andrea clocks such as 11/12MHz clocks with doublers, but not as low as Andrea 5/6MHz clocks with two doublers each.
At this point in time at least, the IanCanada clock is distinctly more blurred sounding than AckoLabs. Also Ian clock is less detailed in terms of instrument textures (e.g. natural piano chord textures), and not as precise is lateral spatial localization. These perceptual results seem more or less in proportion to measured close-in phase noise performance numbers, IMHO of course.
Would I recommend the Ian clock? Yes, to me its definitely better than Accusilicon, and its pretty pleasant to listen to. The downside is that the sound is still missing some of what the Acko labs clocks deliver into this particular system. That said, I would consider the IanCanada clock as probably acceptable for a somewhat budget limited yet pretty high end system (maybe good enough for an improved Marcel dac, say). Probably not desirable for a "cost is no object" level system though. Maybe that proposition can be tested later by driving the Marcel dac with the Andrea FIFO and Andrea/Acko clocks. Could be the Acko clocks and FIFO buffering will offer some improvement in SQ.
Regarding cost, still don't know what it would cost to build a clock board. Do know it would depend on how much of it was populated depending on its intended use. Thus, can't compare cost differences of different clock options at this point.
From this point, I will give the IanCanada clock some more time to see if it settles in any more over the next few days. Eventually, it will go into my build of a Marcel dac board setup with external clocking, and reclocking/isolation of I2S signals. More on that in maybe the next week or two as things progress.
Also regarding the clock board itself, it should be possible to do some tests later to see to what extent the board affects the sound of clocks mounted on it. Previous experiments with a similar clocking arrangement led to the best judged SQ of clocks used in earlier dac projects. Several variations of buffing, powering, bypass, etc., were tried. Based on that experience, so far I'm pretty happy with how clock board turning out.
Clock board is driving an Andrea Mori FIFO board which then drives the dual Andrea DSD dacs.
So far the IanCanada clock has been A/B compared to the Accusilicon 45MHz clock using the same clock board. For the purpose of driving the Andrea FIFO board, the clock frequencies are divided in half using NC7SZ74K8X flip flops. The division is expected to improve phase noise by about 6dB. (45/49MHz clocks are used in order to provide the option of buffered outputs for external clocking of I2SoverUSB.)
The clock board 22/24MHz outputs were also compared with AckoLabs 22/24MHz clocks for driving the FIFO board.
Basically at this point the Ian clock sounds significantly less "blurry" to me than the 45MHz Accusilicon clock. Definitely a step up in that regard.
The SOA AckoLabs clocks were chosen for comparison mainly because they are small and simple to connect/disconnect and move out of the way as needed. AckoLabs measured phase noise is similar to Andrea clocks such as 11/12MHz clocks with doublers, but not as low as Andrea 5/6MHz clocks with two doublers each.
At this point in time at least, the IanCanada clock is distinctly more blurred sounding than AckoLabs. Also Ian clock is less detailed in terms of instrument textures (e.g. natural piano chord textures), and not as precise is lateral spatial localization. These perceptual results seem more or less in proportion to measured close-in phase noise performance numbers, IMHO of course.
Would I recommend the Ian clock? Yes, to me its definitely better than Accusilicon, and its pretty pleasant to listen to. The downside is that the sound is still missing some of what the Acko labs clocks deliver into this particular system. That said, I would consider the IanCanada clock as probably acceptable for a somewhat budget limited yet pretty high end system (maybe good enough for an improved Marcel dac, say). Probably not desirable for a "cost is no object" level system though. Maybe that proposition can be tested later by driving the Marcel dac with the Andrea FIFO and Andrea/Acko clocks. Could be the Acko clocks and FIFO buffering will offer some improvement in SQ.
Regarding cost, still don't know what it would cost to build a clock board. Do know it would depend on how much of it was populated depending on its intended use. Thus, can't compare cost differences of different clock options at this point.
From this point, I will give the IanCanada clock some more time to see if it settles in any more over the next few days. Eventually, it will go into my build of a Marcel dac board setup with external clocking, and reclocking/isolation of I2S signals. More on that in maybe the next week or two as things progress.
Also regarding the clock board itself, it should be possible to do some tests later to see to what extent the board affects the sound of clocks mounted on it. Previous experiments with a similar clocking arrangement led to the best judged SQ of clocks used in earlier dac projects. Several variations of buffing, powering, bypass, etc., were tried. Based on that experience, so far I'm pretty happy with how clock board turning out.
Last edited:
Its an LT1763, but yeah, that's the load resistor for the main regulator. There is also an optional auxiliary regulator that can be used for some things or not.
Why not LT3042/45 ? Guess it could be, but I tend not the like the sound of them. LT1763 sounds okay (with an external NPO noise cap). I like TPS7A33 too.
I try to get away with using only one regulator for clocks and buffers. How power is routed inside the board is different for the clocks (considering the current loops and PCB inter-layer capacitance, etc.), but experiments showed one regulator sounded better than multiple regulators for most of the circuitry on the clock board.
Why not LT3042/45 ? Guess it could be, but I tend not the like the sound of them. LT1763 sounds okay (with an external NPO noise cap). I like TPS7A33 too.
I try to get away with using only one regulator for clocks and buffers. How power is routed inside the board is different for the clocks (considering the current loops and PCB inter-layer capacitance, etc.), but experiments showed one regulator sounded better than multiple regulators for most of the circuitry on the clock board.
Last edited:
Thanks for the reply ... Personally I haven't listened to SMA cables in a structured way yet - the setups more or less have mandated the flexible cable versions - but even so when I have been able to use various cables I have noticed slight differences here and there. Look forward to delving deeper into this hopefully in not too long.
Cheers, Jesper
In other news, there have been some PM discussions with a few people about PCM2DSD modulator versions, and which one sounds best. Although there is a thread just for the PCM2DSD board, it seems to be more of a "how to build this project and hook it up" type of thread, whereas most of the FIRDAC and modulator technical/theoretical discussion has been more centered in this thread. That being the case, and if its okay with Marcel, I would like to summarize some of the various observations about PCM2DSD algorithms and how that fits in with modulator modulator design/theory here in this thread. Also, I think the subject is probably germane to this thread anyway since we need a DSD modulator to get the most utility out of Marcel's RTZ FIRDAC design. Marcel?
Last edited:
Okay, will proceed here. What happened yesterday is that someone contacted me by PM to report an opinion about which PCM2DSD FPGA code he preferred. In addition I kind of summarized how I view the perceptual differences in some PM messages to a few people, so thought maybe its time to say a little more in the open forum about what is perceptually different about the two most recent versions of FPGA code, and to see if there are any ideas as to how to go about solving the optimization problem, which if I understand correctly mostly has to do with modulator characteristics.
If my numbering is correct, the most recent FPGA version for PCM2DSD is v04, and the prior version which I prefer is v03. I compared them with each other and with an arbitrary HQ Player version using AMSDM 512+fs and the Gaussian XL filter to produce DSD256/44.
Basically the main points I have about each one are:
v03 - Good dynamics, open sound (good space between the instruments), best L/R imaging, most distorted/rough
v04 - Smooth and low distortion, closed-in (relative to v03, instruments more blended together), compressed dynamics,
HQ Player - Good dynamics, open-ish sound, low distortion and smooth
To explain a little more, v04 made my system sound like pretty good hi-fi reproduction. v03 made my system sound more like a live music event was occurring in my living room. In that sense v03 sounded more real to me.
In another sense v04 sounded like an inferior, lower performance version of HQ Player; less dynamics, more closed-in. Whereas v03 sounded slightly more open than HQ Player, had clearly better imaging, yet was more distorted/rough sounding.
If anyone else would like to report their listening impressions, I hope they will being willing do so in order that I am not the only one expressing an opinion. More information here will probably better than less information.
--------------------
Moving along, I also recently tried some test versions of FPGA provided by Piotr. IMHO it seems to me like v03, and v04 are local maxima optimizations of particular of particular algorithms. Don't know if Piotr would see it the same way.
What I would like to know from Marcel and others, is how modulator designs are typically optimized for good sound? Is there any rhyme or reason to it?
If my numbering is correct, the most recent FPGA version for PCM2DSD is v04, and the prior version which I prefer is v03. I compared them with each other and with an arbitrary HQ Player version using AMSDM 512+fs and the Gaussian XL filter to produce DSD256/44.
Basically the main points I have about each one are:
v03 - Good dynamics, open sound (good space between the instruments), best L/R imaging, most distorted/rough
v04 - Smooth and low distortion, closed-in (relative to v03, instruments more blended together), compressed dynamics,
HQ Player - Good dynamics, open-ish sound, low distortion and smooth
To explain a little more, v04 made my system sound like pretty good hi-fi reproduction. v03 made my system sound more like a live music event was occurring in my living room. In that sense v03 sounded more real to me.
In another sense v04 sounded like an inferior, lower performance version of HQ Player; less dynamics, more closed-in. Whereas v03 sounded slightly more open than HQ Player, had clearly better imaging, yet was more distorted/rough sounding.
If anyone else would like to report their listening impressions, I hope they will being willing do so in order that I am not the only one expressing an opinion. More information here will probably better than less information.
--------------------
Moving along, I also recently tried some test versions of FPGA provided by Piotr. IMHO it seems to me like v03, and v04 are local maxima optimizations of particular of particular algorithms. Don't know if Piotr would see it the same way.
What I would like to know from Marcel and others, is how modulator designs are typically optimized for good sound? Is there any rhyme or reason to it?
Last edited:
I’ve been swapping out a pair of modules , one with V0.3 the other V0.4 and listening for over a week . I mainly use Amazon prime unlimited with a modded Douk feeding into a HDMi to i2s receiver into the pcm2dsd to Marcel FIRDAC with Bohroks 1632 active filter , I also tried some test tracks sent by Mark .
I’d totally agree with this
v03 - Good dynamics, open sound (space between the instruments), best L/R imaging, most distorted/rough
v04 - Smooth and low distortion, closed-in (relative to v03), compressed dynamics
One thing I’m definitely getting with V0.4 is compressed dynamics . If we could get more of V.O3 dynamics with reduced distortion this would be great .
I’d totally agree with this
v03 - Good dynamics, open sound (space between the instruments), best L/R imaging, most distorted/rough
v04 - Smooth and low distortion, closed-in (relative to v03), compressed dynamics
One thing I’m definitely getting with V0.4 is compressed dynamics . If we could get more of V.O3 dynamics with reduced distortion this would be great .
Just by listening it is impossible to tell what is causing the perceived differences. Distortion can make the perceived sound more dynamic or more compressed. The problems found in measurements of PCM2DSD v03 were not distortion as it is commonly understood.
This type of modulator analysis is not very useful unless the details of the modulator are known. AFAIK currently only the developers of PCM2DSD have enough information for analysis.
This type of modulator analysis is not very useful unless the details of the modulator are known. AFAIK currently only the developers of PCM2DSD have enough information for analysis.
Linear distortion could do it, that is, in relation to the Threshold of Audibility for Group Delay.
In other news, the IanCanada clock has been running continuously for 6-7 days. Looks like its settling in a little more. Imaging is getting a little better, low level instrument textures are getting a little better. As a rough guess, I would say its maybe somewhere in the range of 75%-80% of the AckoLabs clock sound, maybe a little towards the lower end of that range (which is clearly and substantially better than Accusilicon). However, the Ian clock may settle in some more over time, have to wait and see. For now I am going to power it off and mount it on the Marcel dac project wooden base.
For my valve DAC, I just implemented three different sigma-delta modulator algorithms (combined into one circuit that can work in three different modes) and a blind ABCX testing feature (which has never been used by anyone, as far as I know). The modulators all suppress idle tones around fs/2 to some extent; I knew from theory as well as from experience at work that those idle tones very often cause trouble in some way or other.
One modulator algorithm is a choatic single-bit modulator in accordance with Lars Risbo's PhD thesis, that is, with two all-pass sections in its noise transfer function, one causing chaos for low frequencies and one around fs/2. It reduces the idle tones, but does not suppress them entirely. At least I still see peaks in the DFT around fs/2 (simulated with a Pascal program).
The other two are sigma-delta modulators with an embedded pulse width modulator, the ones I called quasi-multibit before. The quantizer of the sigma-delta modulator is a multibit quantizer that is dithered in accordance with nonsubtractive dither theory. The pulse width modulator is not entirely linear, they never are, but by letting the loop filter run at the PWM clock rate rather than the quantizer rate, the nonlinearity is suppressed by the sigma-delta loop. On top of that, I randomly rotate the PWM patterns. The advantage of that is explained in my valve DAC article, the disadvantage is that the output becomes less suitable for NRZ DACs.
Theoretically, the chaotic modulator has the best noise shaping due to having the highest quantizer sample rate (largest oversampling), but the other two have far better suppression of idle tones around fs/2 (in fact, I didn't see any such tones in the DFT).
In measurements, the chaotic modulator had weird non-stationary noise issues when playing silence, and the quasi-multibit mode with the largest number of quantization levels (my PWM8 mode, I later made a PWM9) had the lowest audio noise floor. This is related to an interaction with imperfections of the valve DAC, presumably an imperfect settling issue. My CMOS logic gate DAC has nearly the same audio noise floor with all three algorithms.
Of the few people who built a valve DAC including the digital part and listened to it (informally and not blindly), most also preferred the PWM8 mode. Hans has the same experience with the logic gate DAC I once lent to him.
Last edited:
It's a bit more complicated than just simple algorithm analysis...
I'll try to describe it.
At the end of work on version v3, Mark listened many beta versions.
In these beta versions the algorithm was not changed, the level of dithering too was not changed.
We only checked at which input signal level the dithering should work.
Of course, we have previously analyzed the ranges in which we should to do this.
Differences in this level were sometimes only 0.04dB (this is crazy) and the differences in sound were very large.
There were also differences in measurements, but probably not that big.
It couldn't be said that increasing or decreasing this level, for example increases soundstage or something else...
It was rather unpredictable...
We don't really know what influences the sound.
We can, for example, reduce/increase distortion, but we don't know how it will affect the sound.
But we're not the only ones who don't know it
But of course there is a solution for it.
You need a lot of time, patience, a good system etc. and we can make different versions of the firmware and you can measure/listen and maybe we will be very lucky and make a better version...