Interface juggling and function topology adventure.

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
This thread aim to challenge the usual way of function allocation into boxes and their typical physical position wrt. different interfaces.

This had to end up in EE as there is no suitable sub-forum to discuss system solutions... maybe something to add?


//
 
Last edited:

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
I'm awaiting some posts to be moved here. Hopefully it gets clearer ;)

Meanwhile I'll add the below drawing... (SB: e.g. Squeezebox)

//
 

Attachments

  • topo1.png
    topo1.png
    151.3 KB · Views: 141
Last edited:

TADIGITAL

Member
2016-03-01 2:19 pm
I think I got where our mismatch was, see the modifications to your pic.

The pwm generator needs additional control input through I2C. This input controls the general items like how to generate the pwm, what filter settings, volume, ...
As we consider pwm as analog the wiring should be short, very short! Optical transfer may be possible but will be very delicate - Implementing this is ambigous and hardware usually used for TOSLINK may not be sufficient (but maybe a interesting approach to try!).
Audio Input to the pwm generator usually is I2S which can be generated from any type of source with a small electronic (DAC (e.g. TI), spdif/toslink --> i2s converter (e.g. Wolfssohn), ...).

Eventually the controller and the SB (which I like a lot) may go into the same device. Many "high-end" people do not like TOSLINK (and SPIF) due to timing - we used I2S for this which worked out excellent - with many more components.
 

Attachments

  • topo2.png
    topo2.png
    111.5 KB · Views: 124

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
The pwm generator needs additional control input through I2C. This input controls the general items like how to generate the pwm, what filter settings, volume, ...

Yes, I didn't tdraw the control plane, only the "music plane".

As we consider pwm as analog the wiring should be short, very short!

Your analog path will be longer than what I sketched as the analog path also includes the speaker wire.


Optical transfer may be possible but will be very delicate - Implementing this is ambigous and hardware usually used for TOSLINK may not be sufficient (but maybe a interesting approach to try!).
Audio Input to the pwm generator usually is I2S which can be generated from any type of source with a small electronic (DAC (e.g. TI), spdif/toslink --> i2s converter (e.g. Wolfssohn), ...).

Eventually the controller and the SB (which I like a lot) may go into the same device.

Agree!

Many "high-end" people do not like TOSLINK (and SPIF) due to timing - we used I2S for this which worked out excellent - with many more components.

There are now much higher performing opto systems that are feasible cost wise than when Toslink was conceived.
 
Last edited:

TADIGITAL

Member
2016-03-01 2:19 pm
Hi TNT,

I don't see the pwm being longer - that's exactly why we set it up that way. The wiring from the audio power supply to loudspeaker (+-50VDC give approx. 300W audio power) is interrupted by the Mosfets to switch on/off. This is the same for every digital power stage.
The control line(s) for the Mosfets (which end up at the gate(s)) are pwm, and its favourable if they switch very high frequency and do this very accurat. Therefore these (low power) wires need to be short. We put the pwm generator (green box) close to the power stage exactly for this reason.

The other architecture option would be to send the pwm to the power stage over wires from the remotely placed controller and pwm generator. This requires to have the delicate high frequency mosfet control lines with long wiring (optical would be a excellent choice if timing accuracy and temperature dependency issues can be overcome). There are a few other control lines (like power on/off, reset, overtemperature control, ...) which are easier to handle. The smart idea here is that the power stage has fewer components and gets smaller.

So this leaves us with the two options:
a) power stage close to pwm generator
(+ short pwm wiring, - one pwm generator close to power stage(s), but up to 8 in one box)
b) power stage distant to pwm generator (+ fewer components at the power stage, - high frequency and timing accurate wires are long, + only one pwm generator per system).

I like the idea of to use a optical pwm interface but I'm not sure it can be done that easiy (I guess we should have << 1nsec control, which would require > GHz bit rate...
If somebody has a solution for this I'd like it.
 

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
Hi!

The timing issues in the pwm generation is interesting. I don't see anything about the sawtooth accuracy generation and pwn coding and if oscillator jitter is important? Is pwm immune of jitter? Would not think so.

So why don't we just move the whole class D stage into the "Unit"? That would be my red version in post #3.

But agree, mowing the pwm signal over opto would save space and make the position of the box containing the pwm generation very flexible.

One can apparently do 140m 100G over OM4 - https://en.wikipedia.org/wiki/Multi-mode_optical_fiber

The OM4 fiber is cheap but how about receiver and transceivers? I suppose a brutal layer 1/1,5 is to be used and not ethernet etc higher level coding...

//
 
Last edited:

TADIGITAL

Member
2016-03-01 2:19 pm
The timing and jitter probably is not a big issue if you use the same clock for digitizing and for converting back into the "high power digital" world. In this case they should cancel out. But if they are not the same (or if you use digital music sources) it matters a lot.

(the sawtooth picture only holds if you use an analog signal and digitize it - which gives you double noise and distortion sources).

If we think for the pwm over opto we cannot have any protocol - this is the time domain of the audio signal (or the true "analog pwm") and needs to be transferred without losses. Has anyone experience with the TI PWM concept? I guess each of the pwm signals (3.5µsec @ 280 kHz) needs to be digitized with 16(24) bit?? So this ends up with a timing of 54psec(213fsec) which is far out of reach for a diy project (and for the analog path this sounds hard to meet as well) - can anybody here in the forum tell me where my mistake is - I'm sure something is wrong with these numbers

...
And so we are back to the original setup with the pwm generator close to the power stage. I still separate these into two building blocks, as the requirements for the pcb, wiring and signal quality are totally different for these two. We can still have them side by side which then gives the flexibility I proposed in the early posts and which avoids the issues discussed above.

Regarding the TOSLINK and the jitter issues - you are right nowadays optoelectronics is far better than it used to be at the time TOSLINK was defined. But it's not only the jitter in the physical layer which the "high end" people dislike. The protocol has some inherent jitter due to the metadata coded in the spdif. This was the reason we used I2S because here jitter is non relevant due to the input gates at the receiver. One can improve that even further if the MCLK is utilized (which is not part of the I2S spec) and further if this is based on a good clock... .. all this is additional margin to improve further.
 

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
The timing and jitter probably is not a big issue if you use the same clock for digitizing and for converting back into the "high power digital" world. In this case they should cancel out. But if they are not the same (or if you use digital music sources) it matters a lot.

(the sawtooth picture only holds if you use an analog signal and digitize it - which gives you double noise and distortion sources).

If we think for the pwm over opto we cannot have any protocol - this is the time domain of the audio signal (or the true "analog pwm") and needs to be transferred without losses. Has anyone experience with the TI PWM concept? I guess each of the pwm signals (3.5µsec @ 280 kHz) needs to be digitized with 16(24) bit?? So this ends up with a timing of 54psec(213fsec) which is far out of reach for a diy project (and for the analog path this sounds hard to meet as well) - can anybody here in the forum tell me where my mistake is - I'm sure something is wrong with these numbers

Why digitize it? Just send it as light on-light off corresponding to when the flank shall be high or low. PWM is just switching between low and high and its the time *when* it switches that carry the relevant info - right?

Pwm_duty_cycle.gif


...
And so we are back to the original setup with the pwm generator close to the power stage. I still separate these into two building blocks, as the requirements for the pcb, wiring and signal quality are totally different for these two. We can still have them side by side which then gives the flexibility I proposed in the early posts and which avoids the issues discussed above.

And I suppose in one and the same box and even on the same circuit board I assume?

Regarding the TOSLINK and the jitter issues - you are right nowadays optoelectronics is far better than it used to be at the time TOSLINK was defined. But it's not only the jitter in the physical layer which the "high end" people dislike. The protocol has some inherent jitter due to the metadata coded in the spdif.

I think this is not really inherent jitter but a problem for the decoder and clock extraction circuit not to be influenced of some repetitive patterns which per definition is not intended for carrying info about time. An implementation design flaw and not a specification flaw - this has been mostly cured nowadays when the industry has accepted and understood the relevance of time accuracy at the point for D/A conversion. But in practice when sp/dif where used as a time reference it is (was?) a problem.

This was the reason we used I2S because here jitter is non relevant due to the input gates at the receiver. One can improve that even further if the MCLK is utilized (which is not part of the I2S spec) and further if this is based on a good clock... .. all this is additional margin to improve further.

Isn't i2s is a short haul protocol intended for "on circuit board use" - sp/dif for external consumer interface so not direct comparable. Its often advised against using I2C for longer runs or over connectors between boxes.

Still you have not made it clear to me why you don't seem to find a one box / one channel solution interesting. Why is this? Or maybe you did :)

"And so we are back to the original setup with the pwm generator close to the power stage. I still separate these into two building blocks, as the requirements for the pcb, wiring and signal quality are totally different for these two. We can still have them side by side which then gives the flexibility I proposed in the early posts and which avoids the issues discussed above."

Reading it again I realized you did so You are OK with 2 boards inside one and the same box - right?

//
 
Last edited:
Hi TNT,

There is no need to digitize the PWM. It was argumentation to show the timing requirements, which are very tight for the "analog pwm" signal. I would be happy if somebody can show where the calculation above is misleading - or confirm it.
(Full Signal corresponds to 100%PWM, 50% Signal to 50% PWM, therefore I guess the time to switch high-low/low-high has to be accurate to the bit depth of the audio signal - and this is a requirement which is hard to meet!)

Optical fibre to couple the pwm - this is a really smart concept! But - again the timing is critical and a deep understanding of temperature and part to part tolerances for optical transmitters and receivers is needed to implement it without degradation in quality. Does anybody know of such a product? I mean the amplifier would then be a MOSFET with optical input - nice and simple amplifier!

If you look at differential signalling components (e.g. LVDS) we might get closer with the required accuracy for the switching timing (LVDS achieves ~nsec).

Yes, we put different functionalities on different boards. Having two/three/four/... boards inside a device is not a problem and allows flexibility - for users and designers.

One example for this is I2S input. A building block with I2S input and a I2S output connects to the pwm generator. It is easy to set up a component with TOSLINK input and I2S output, so the user can choose what to implement - without redesigning (and producing) the whole device.

I agree I2S is not industry standard as interconnect of the devices - but there are a few suppliers which allow I2S inputs. And it avoids the timing issues (which you described correctly) with the TOSLINK and spdif transmitters. Utilizing is not for free, more wires are needed (I2S+ MCLK = 4 lines, TOSL/spdif = 1 line!). Pure I2S is intended for short length IC comunication. But as the signals are separate into 3(4) lines it can easily be sent over a CATx wire or one HDMI cable (which leaves a few spare lines for additional information). Of course single ended transmission of the MHz I2S signals is not a good idea, a LVDS transmitter/receiver combination can be used in between (or a different technology allowing to transmit the frequencies).

Disadvantage of this concept is that control lines need to be at the pwm generator (dsp). Also haveing one dsp/pwm-generator board in each of the "units" may not be the most cost-effective solution.

I'll prepare a pic of the functional blocks and post it here later today.
 

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
Hallo!

Yes - timing is the challenge. This is why I would strategically like to stay in PCM as long as absolutely possible. As long as you maintain the PCM words you are 100% good. This is very close to the speaker driver. Now there is one challenge with this and it is again time. Making the PCM conversion close to the speaker driver mans that this process will be exposed to high sound pressures which can/will effect the clock with drives the D/A conversion. The very nice -116 dBc at 10 Hz phase noise plots that made you click "Put in basket" is to waste as you shake that oscillator.

Time will come back and bite you eventually and this is why a system architecture is need to advance the art rather than performing OPamp rolling and red boutique capacitor shining.

So in my mind the ideal topology pushed out the D/A which is followed very close with the power stage and the timing/clock is injected into the D/A over fiber so that the oscillator can be put in a quiet place - given that time precision can be maintained over fiber in a better way than having the oscillator locally. Maybe it isn't a big problem but do a spectral analysis of a clock and hit it lightly with a pen.... This probably takes some theoretical analysis as well as some practical tests - Im sure something happens (dispersion?) if you squeeze a fiber cable so it is most probable not immune to the other crude world.

Looking forward to your pictures TADIGITAL.

//
 

TADIGITAL

Member
2016-03-01 2:19 pm
Attached a rough block diagram.

The two green and the red are close to the loudspeaker, logic controller to be decided from system standpoint.

Using TI there are a few choices for the pwm generator:
PWM Modulators | Products | TI.com

And they offer a variety of matching Mosfet H-bridges with lots of different power and quality ratings:
Class D Amplifiers | Products | Audio | TI.com

I know this sounds a bit like a TI ad...still our listening experience for that customer was pretty nice (TAS5508 and high power outputs)..

I suggest for a family of DIY boards to forsee galvanic isolation for the I2S input to the pwm board and for the I2C signals from logic control. This allows to separate audio relevant electronic and also to minimize damages. To be decided which isolators to use and where exactly to put (input of pwm board or input of audio input board). Disadvantage may be the extra power supply for the logic...

Sound quality and power depends mostly on the red box's design.

...

TNT, I don't get it why your PCM coding should avoid the timing issues? You still need to have the timing accurate to 16/24 bit, or do I not get your point?
 

Attachments

  • uni-IIa.png
    uni-IIa.png
    86.7 KB · Views: 53

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
Hi!

I must say I thought that the Class D chips where "complete" and didn't need any PWM modulators... Now realizing that one can either build one by 2 components as per your component suggestion or as a 1 component solution e.g. TPA3251D2 | Mid/High-Power Class D Amplifiers | Audio | Description & parametrics

I think I made the suggestion that while a signal is coded in PCM no timing error can be introduced - only at the point of D->A conversion can jitter be imposed. So granted no bit errors the signal is intact as long as in PCM. That was my point and insight for my idea of topology. At the location where D->A conversion is performed a accurate clock is needed for sure. I want to carry the signal as far out in the chain as absolutely possible in PCM format to minimize errors - errors happening in D/A conversion and the following analog domain.

Looking at the supplied picture I deduct that if the speakers are 3 meters apart, the smallest length of analog signal path is really 3 meter. So you need 3 meter of 1000USD/meter cable to make it sound good? :)

But I want to have only 10cm analog path length :bawling: :)

Would you say that typically a 2 component (PWM mod + H bridge) is better than an integrated solution like tpa3251d2 etc?

//
 

Attachments

  • syst.png
    syst.png
    56 KB · Views: 49
Last edited:
TNT,
you have quite detailed requirements regaring the size of the unit :)...

The 3 m is not true as you can set up the device with one pwm modulator per channel as well (You don't need to have only one in the system!). As the timing from digital input signal to pwm output is determinisitc (no buffering as in the audio controller) is performed - and in this case you simple have to "copy" the digital signal - which can be done without loss (see attached).

I didn't check for the physical size of the boards but other than that your minimal analog signal path length would be met with my concept.
 

Attachments

  • uni-IIb.png
    uni-IIb.png
    117.1 KB · Views: 41

TNT

Member
Paid Member
2003-04-26 10:25 pm
Sweden
OK , I'm with you!

Can you package that into some physical boxes please :)

My drawing program apperently likes decimals ;)

Looking at my Unit drawing the ethernet + toslink docent make sense really..... ethernet cables are awkward in numbers.. on theater hand toslink can't carry control info. Maybe to put all DSP and level control in the central unit and let the Unit have fixed gain.

//
 
Last edited:
... missed to mention that the tpa325 requires analog in - with all the losses inherent to analog and with the extreme long analog path you get from your audio controller...

to be honest - I never compared a well designed analog-digital converter chain to a full digital chain with similar quality (design, components, manufacturing) so I cannot tell you. I listened to quite a few (different) analog A and AB amplifiers and I listened to well done pure digital amplifiers. A comparison with the same room, loudspeakers and audio material and with my own and known music material is something I would really like to do - but what I cannot afford to do (costwise and timewise!), not to mention the risk that I may end up spending lots of €€€€...