Carte allo digione signature - essais et comparatifs

[Roberto]

HS ON

J’ai trouvé cette intervention de la part de John Westlake co-designer de la Project Streambox S2 Ultra sur un forum anglophone. Je trouve interessant pour comprendre les efforts pour optimiser cote hardware une simple carte Rpi à des fins audiophiles. C’est valable pour Allo, Metrum et bien d’autres...

HS OFF



As a co-designer (with Jarek) I can help answer some questions about the ProJect Streambox S2 Ultra - ProJect is our 3rd OEM “customer” for our streamer solution undefined

The roots of this design is based on our own internal requirement for a streaming solution for a crowd funded DAC I've been developing for the past few years. The crowd funded group consists of about 150 development sponsors – so I have to be creative to maximise the limited funding resources.

One requirement for the DAC was for a Streaming solution that allowed HiRes Bit Accurate (no data manipulation), DSD support and allowed the DAC to be Audio Clock master. With our limited 150pcs to 200pcs DAC production – nobody was interested in supplying a streaming solution so I was forced by such circumstances to turn to an open source software.

The Open Source community is very active in this area, but the hardware solutions are based on non optimised off-the-shelf computer hardware. While the Raspberry Pie is a great low cost computer hardware – however it’s been designed with cost as the primary objective and understandably not with audio optimisation anywhere near an important consideration.

So while there is zero chance we have the resources to develop / support the streaming software for our small volume group sponsored project, we can develop the hardware with our eyes closed shut.

It became apparent that Volumio offered the best feature set (just missing SBT / LMS support) – but as an old school turntable / CD user I struggled with its UI. I needed to find a way to have a customised version of Volumio, but with only 150 to 200 potential units there is no way we could afford to directly pay for the customised software development.

So, as our requirements where rather desirable for HiFi manufactures (Bit Accurate HiRes (no data manipulation), DSD support and allowed the DAC to be Audio Clock master) I developed a plan to offer our streamer hardware solution to a few select customers I’d worked with in past (who all struggled to find a decent streaming solution).

Now with a rather large carrot (the potential of large OEM sales) – I was able to convince Volumio’s founder Michelangelo that he could make a nice little business from developing a custom OEM version of his software that would work with our hardware platform – in exchange we would get a supported software solution for our 150 – 200 group funded DAC undefined

After several false starts (BeagleBone, and a few other cheap Chinese SBC’s) we about gave up on the streamer side of the project when the raspberry Pie foundation released the long promised Quad Core industrial version of the RPi3. This industrial version of the RPi3 called the CM3 module allows us to optimise the streamer hardware (PSU, RF and Clocking) while maintaining software compatibility with RPI. An important goul of our streamer solution is that we wanted an optimised “audio” streaming solution while not limiting the customer to a single software solution – for those skilled in software they can install any of the RPI based media players (Moode, LMS and even Kodi) etc.

Another important advantage of the RPi based CM3 module is the use of onboard eMMC memory thereby removing reliability issues (and speed) when using “External” SDcards.



In answer to “why the extra cost” over the standard RPi3 – it’s required to give a basic explanation of the RPi3 weakness WRT high end audio applications (remember is designed as a lowest cost computer platform NOT a highend audio streamer). These weakness in no way are a criticism of the RPi3 – but rather its design limitation from my perspective as an audio designer.


Basically its weakness are all based on Clock Phase Noise (Jitter) and local PSU noise (which directly impacts the Jitter performance) and also the resultant local RF emissions:-



1. The USB HUB / LAN IC is clocked by the CM3 SoC which has significantly higher jitter then an external Clock oscillator. More detrimental is that fact that clock phase noise is heavily correlated with the SoC software processing – resulting in Data correlated jitter.



2. The USB HUB / LAN IC PLL clock circuits are powered from the switching supply rail that also powers the SoC Memory, so as data is “processed” though the memory this causes modulation of the PSu Rail directly correlated with the Data patterns which then modulates the critical PLL in the USB HUB / LAN which is used as our USB “source” via this noisy PSU rail.



3. As one would expect little care has been taken with the quality of the 5V rail for USB DAC power – in fact its really pretty poor by “Audio” standards.




4. There are multiple “free running” switching regulators on the RPi PCB – whose clocks are not related to the USB clock Rate so there switching products are free to induced multiple / complex RF and Ground noise spurie in the system.



I could list more weakness, but you can start to appreciate the “Deficiencies” (not unexpected) of the simple Rpi3.



1. The little ProJect streamer tackles these issues by making the USB 24MHZ clock the system “Clock master” and cleanly generating the LAN’s 25MHz clock and the multiple local switching supply’s operating from a divided down synchronised frequency of this 24MHz.



2. Great care has been taken with the internal PSU’s – I count 8 low noise PSU’s of the top of my head.



3. The USB host signal from the CM3 is RF filtered and USB “Detoxed” (the poor Phase noise of the CM3 USB Host controller) is attenuated by two Cascoded USB Hub bridges (and RF filter) resulting in a much attenuated USB Jitter on the “DAC” USB output.



4. The DAC 5V feed has a very low noise regulator – whose voltage and current can be monitored via the streamer software.



5. The DAC 5V USB voltage output can be disabled via software for DAC’s that don’t require 5V.



6. An external USB bypass to the DAC USB output is proved, this bypass “Detox’s” the external USB signal in the same manor as applied to the RPi CM3 USB Host output.



7. 16GB onboard eMMC is available for OS / software & local files.



Sadly, all this makes for an expensive hardware solution – with “low volume” production costs are factored (ProJect will manufacture thousands Pcs – not millions) – the end selling price is just a fact of European low volume manufacturing – this is not China manufactured mass market pricing…


I don't seem to have an image of production versions of the PCB, but here's an early prototype of the board - where you can see its not simply a RPi3 in a smart box  undefined


Dropbox - MiniStreamer A01 PCB.JPG

Sorry for the errors in the published documentation / marketing – no text is run past me for proof reading…