PCIe USB and SATA card modified for audio

[numlog]

Here is modified ~$30 SATA and USB3.0 PCIe cards that Im using with really great results:

 

Both are powered externally via the PCIe power traces which requires some delicate modification of the card.  
The USB card in the back has a SATA power input that supplies 12V but this is completely useless for this purpose, it only supplies power for the USB ports, not the USB controller chip and other ICs on the board. Not only that but its just supplementary power, the card works 100% without any SATA power using 12V from the PCIe slot.

 

The external supply is using these cheap TPS7A4700 regulators to provide 3.3V for both cards and works excellently.

The SATA card has a significantly higher constant current draw than the USB card so better heatsinking is needed, the bulk of USB card current draw will be from whatever device you connect to the ports, ideally you shouldnt be using 5V from the card to power your DAC.

The SSD has its own external 5V supply too.

 

So what is better than a $500 dollar silver plated SATA cable to connect your drive? none at all.

You just need 2 of these bare SATA connectors  and good quality solder (I used silver) to make a tiny SATA to SATA adapter, extremely inexpensive and extremely impractical for SATA performance that you simply cant buy.

Its extremely delicate too so connecting and supporting the drive felt like it could be an issue but I got very lucky. When the 2.5'' drive is resting sideways on the 16X PCIe slot its SATA slot is exactly the same height as the SATA slot on the card, the SSD is as a light as feather which helps too. Obviously this is NOT a portable solution but it couldnt have worked out more perfectly.

Even completely unshielded it is pure and transparent compared to all other SATA cables, Though if this is like what you get with a $500 dollar SATA cable I would be kind of disappointed.

 

The SATA card's performance really surprised me. With the USB card it didnt really sound much different to the USB3 ports on the MoBo, it was just cleaner/smoother as you'd expect with clean external supply, but with the SATA card it wasnt just cleaner, there was a very clear jump in performance.

 

One final thing is that I compared the USB3.1 port on the MoBo with other USB3.0 ports on the MoBO and with the modified USB3.0 card ports, what it showed is that even for USB2.0 devices USB3.1 is clearly better than 3.0, it's simply more realisitic than either the MoBo 3.0 ports or my card 3.0 ports, with each version having its own sound character.

To confirm this I have ordered a USB3.1 card from the same manufacturer (Startech) to make the same modifications and fairly compare to the 3.0 card.

If the 3.1 card turns out to be better than 3.0 card its possible a cheap modified 3.1 card could outperform the expensive Audiophile USB3.0 cards on the market.

 

 

[rickca]

11 minutes ago, numlog said:

If the 3.1 card turns out to be better than 3.0 card its possible a cheap modified 3.1 card could outperform the expensive Audiophile USB3.0 cards on the market.

I'm using this Startech 4-port dual channel USB 3.1 Gen 2 card and it sounds great.  Several other CA users have reported that they really like it as well.  Startech also makes a 2-port single channel version of this card.

https://www.amazon.com/StarTech-com-USB-PCI-Card-USB/dp/B071DFQ6TW

[sandyk]

Numlog

 

Very interesting.

[numlog]

3 minutes ago, rickca said:

I'm using this Startech 4-port dual channel USB 3.1 Gen 2 card and it sounds great.  Several other CA users have reported that they really like it as well.  Startech also makes a 2-port single channel version of this card.

https://www.amazon.com/StarTech-com-USB-PCI-Card-USB/dp/B071DFQ6TW

A 2 port (1 internal , 1 external) 3.1 card was on sale for 15GBP, good if you only need one port but I think I got the last one though. 

 

edit:yeah unfortunately not on sale anymore

 https://www.amazon.co.uk/StarTech-com-Port-10Gbps-Card-Type/dp/B01AY7U8NC/ref=sr_1_5?ie=UTF8&qid=1537312516&sr=8-5&keywords=startech+3.1+2+port+card

[numlog]

The SATA connector is pretty ridiculous in hindsight, reduce electrical distortion at the cost of creating extreme impedance mismatch, im surprised it works but it does, it sounds decent enough too.

\I think most cheap SATA cables arent made to spec anyway, as they will still work in the end even when grossly out of spec. The best sounding sata cable I found was an amphenol one, ASUS cable were also good but the DIY connector somehow managed to sound more resolving, which the amphenol proved to me was real resolution and not an effect of impedance problems, the amphenol also shows the weird effects the DIY connector was creating.

 

same thing with DIY USB cables, extremely short silverplated USB cable, soldered directly to usb port and input sounds better than any generic cables (some of which are very good) but switching to a 0.7M supra was best if not slightly less resolving.

Seperately shielding data lines in USB also improved resolution, its hard to know if it would have this effect with a properly speced cable too.

Im interested in attempting to build a USB cable that tries to be closer to spec, using 30AWG UPOCC silver and keeping around 10CM in length should give put the impedance in the right ballpark going by online calculator, will build a few to try seperately shielded data and soldering directly to ports  and find out if they matter

[ingemar]

This is what i have been looking for,can you explain how you did it with the sata card?

Kindly/Ingemar

[numlog]

find the 3.3v power pins , you can use this as guide http://pinoutguide.com/Slots/mini_pcie_pinout.shtml

 

the traces from power pins on my card look fairly distinct and werent difficult to identify, they are fatter than data and other non power trace, are grouped together and only go a few mm onto the board where it connect to internal PCB layer through several vias. The ground pins all connect to ground plane which is easy to see.

use a sharp blade to cut the power pins

 just as it enters the solder mask, try not to use much force.

(only power, do not cut ground pins, they need to be connected to PC to work)

 

again use a sharp blade to gently scrap away an area of solder mask on the power traces to solder onto, 

do the same for the same for an area of the ground plane (maybe do ground first to practice).

 

[ingemar]

4 minutes ago, numlog said:

Perfect,thanks a lot!!

 

 

 

find the 3.3v power pins , you can use this as guide http://pinoutguide.com/Slots/mini_pcie_pinout.shtml

 

the traces from power pins on my card look fairly distinct and werent difficult to identify, they are fatter than data and other non power trace, are grouped together and only go a few mm onto the board where it connect to internal PCB layer through several vias. The ground pins all connect to ground plane which is easy to see.

use a sharp blade to cut the power pins

 just as it enters the silkscreen, try not to use much force.

(only power, do not cut ground pins, they need to be connected to PC to work)

 

again use a sharp blade to gently scrap away an area of silk screen on the power traces to solder onto, 

do the same for the same for an area of the ground plane (maybe do ground first to practice).

 

 

[numlog]

Replacing the clock on these cards with something much better (crystek 575) is super easy 

 

In the datasheet for the Marvell SATA card chip it says it can except a 25MHZ clock input or drive a crystal with its own internal circuit.

it uses a crystal, i assume as its cheapest option, removing the crystal and using a cheap but decent 25MHz XO gave noticeable improvement.

 

I also tried with the ASMedia USB3.1 chip which in its Datasheet says it can use a 48MHz clock input or drive a 20MHZ crystal with its own internal circuit (odd that it wants different, non divisible frequencies), it is also using a crystal. sadly using the 48MHz XO input didnt work.

 checked the datasheet again and found in the notesthat apparently the 48MHz clock input feature was '''removed'' in a newer  version of the chip.... very annoying.

it could be possible to use 20MHz XO instead, will have to try it.

[marce]

On 12/3/2018 at 7:11 PM, numlog said:

 

Seperately shielding data lines in USB also improved resolution, its hard to know if it would have this effect with a properly speced cable too.

Im interested in attempting to build a USB cable that tries to be closer to spec, using 30AWG UPOCC silver and keeping around 10CM in length should give put the impedance in the right ballpark going by online calculator, will build a few to try seperately shielded data and soldering directly to ports  and find out if they matter

Cable characteristic impedance is not related to or affected by the cable length... So an online calculator that uses length is wrong.

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

USB is diff pair for data transmission, how will you cater for that... the impedance is 90R differential...

 

[numlog]

Just now, marce said:

Cable characteristic impedance is not related to or affected by the cable length... So an online calculator that uses length is wrong.

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

USB is diff pair for data transmission, how will you cater for that... the impedance is 90R differential...

 

Ive accepted DIY digital cables are a complete waste of time.

I had no idea the complexity of the impedance spec till today after making a thread earlier (not for DIY purposes, to understand how ''HiFi'' SATA/USB cables could advertise use of the same low gauge for all cable lengths)

 

After more time listening to the supra USB cable its obviously much better than any DIY or generic cable, The extreme mellowness compared to the other cables took some mental adjustment, there is plenty more detail and realism present its just not in your face whatsoever. With high quality conductors and lots of shielding you can probably make a USB cable thats sounds better or gives an impression of 'quality'' or ''detail' compared to generic cables, but its still a toy compared to the most basic ''proper'' cable.

[marce]

If anyone is interested in seeing how characteristic impedance is calculated
From the geometry of the signals and their return path the following application
is a good example. Whilst PCB orientated it shows the relationship of geometry 
to characteristic impedance.
http://www.saturnpcb.com/pcb_toolkit/
It also has many other useful calculators for PCB design…
For an approximation of a diff pair in a cable use the Differential Pairs  tab, then use:
Broad Cpld Shld for shielded cable (with shield connected to GND.
Broad Cpld NShld for unshielded twisted pair cables. 
The following screen shots show the effects of different dielectrics, the first an 
approximation of PCV (Dk = 4) the second PP (Dk 2.25) and how the distance
 between the conductors has to be changed to maintain the same characteristic impedance.
This is a nice simple calculator that gives the required figures for most work, 
when it gets serious then the SIV (signal integrity verification tool) uses a 3D 
field solver based of the actual geometry of the PCB, cable manufacturers use a similar 
tool designed to handle the complexities of cable shapes and internal geometry.

 

 

[numlog]

Just want to confirm using a 20MHz XO instead for the USB card worked, you can probably do the same for Renesas uPD72020 cards too which use a 24MHz crystal.

Now, AFAIK the async DAC theory is that PC side USB clocks shouldnt matter, this is very obviously not true.

 

There is a true improvement to sound, affecting resolution and timing, and makes the whole PCIe card endeavor feel worthwhile.

I hope the crystek clocks wont be too severe a case of diminishing returns but they're definitely next on the to do list.

[numlog]

I had bought an M.2 NVMe drive to see how it sounded compared to SATA SSD as an OS drive, the M.2 did seem technically better, it was less musical and less natural (more glare) but the slight smear, fogginess of the SATA in comparison was gone, it was hard to choose between them.

Both were using low noise linear supplies.

 

Really the M.2 wasnt better at all, even its advantage of being a native PCIe device wasnt the explanation to its technical ability. It was simply that the M.2 was more isolated from the storage HDD and PCIe slots.

 

The SATA OS SSD was connected via the PCIe SATA Card along with the SATA storage HDD when comparing it to M.2 SSD using MoBo M.2 slot, this was the issue.

Individually both these SATA drives had sounded best from the PCIe SATA card instead of internal SATA,

Assuming that both of them using the card at the same time would still be best turned out to be wrong.

 

With the SATA SSD using the internal chipset SATA interface and the HDD storage drive using the SATA card the  SQ improved beyond the M.2 drive without the undesirable qualities.

Using another seperate PCIe SATA card for the SSD is an option but Im going to assume that internal chipset SATA provides an extra layer of isolation from the PCIe lanes, which could be important not just for the storage drive SATA card but other PCIe cards too.

 

There is also power supply isolation to consider.

The HDD and SATA card use a low noise supply. the PC sounds better when the OS SSD were powered by the noisier but more isolated PC PSU than the same supply as the SATA card and HDD. One or both are interfering with the other.

 

I was hoping M.2 would be better so buying a highend SATA cable could be avoided but at least since SATA is now truly superior a better cable will feel like an upgrade and less like a bandaid for the sound issues.

 

EDIT: I dont mean to wrtie off M.2 NVMe drives, these impressions could have been different if other M.2 NVMe and SATA SSDs were used .

Samsung 850 EVO 2.5'' SATA SSD and Kingston A1000 PCIeX2 NVMe  SSD were used here - both 240GB.

newer, faster SATA SSDs or NVMe SSDs could be different.

There are also M.2 SATA drives  to try.