Understanding FPGAs

[ZolaIII]

Hi,

Thanks a lot for taking time to reply to me.

I think I begin to get to understand a little more about FPGA's.
Quick question : if it is like those little squares, can it be that a "serious" manufacturer can update it's product (an make that update available to you & me) with a "firmware" update ?

Now, back to the subject : as I hear you, the x8 may not be bad, but you would serioulsy consider something else for the price.
Something that would measure (much) better but will only marginally sound better, if it sounds different at all. And with features like, maybe, balanced.
Personnally, I do have no plans on using long runs of cable so I probably do not need balanced and, the common saying on another thread is that the extra power in balanced would not be need given the used headphones (Oppo PM-1 : 32 Ohm and 102dB).

But of course, if you can advise something that looks good, has about the same price (up to 250/300 dollars) and that has stellar measurements, I may consider your advice (provided the device is available easily in Europe/France).

Thanks again for taking time to reply

Almost every audio device including & little tiny USB dongles has an EPROM which contains firmware and can be updated to resolve some issues with functionality that's already there. Those with FPGA-a can be upgraded regarding functionality & even new functions (new codecs different algorithms and cetera). But don't expect much from manufacturers. In rare cases which I can remember always whose a developer community which drowe the progress & new goodies to play with.

Thers lot of ChiFi mesured hire & with some actual feadback from users regarding how that really works. Most of it is available through Amazon & plenty to choose from for 200~300$. Take a look at Topping and SMSL models mesured and popular hire for instance. Of course priority is what fit's your needs the best. You won't get something like final answer or concrete proposition from me (there's many reasons for that).
For me the SMSL M300 MKII for instance represent a great choice for what it offers, how it measures and that the price right now is only 200 & couple $. Of course other's could disagree with me or simply don't like it, or insist on some future it doesn't have (LDAC, volume potentiometer or something else). Everybody's taste is different.

You are welcome.

 

[PenguinMusic]

At 3:02 in the Chord presentation, you will find a lie where they claim that there technology is necessary for timing accuracy which is not true. All they can claim is oversampling to push the frequency up and reduce changes in phase due to analog filtering. Their claims to improved accuracy only hold water if you ignore analog post filtering. They claim huge numbers of taps, which is good, but ignore the real "processing" that occurs post DAC in the analog domain.

Hi,

Not sure I really understand all you're writing, but I can say that those kind of claims always makes me somewhat suspicious...

Even for the Aune x8 I'm looking for when they claim that it uses their own FPGA that 6 years to develop.
Me, I don't care about that.
How does it measure/sound ?
That's all I wanna know...

I know of colleagues that are working in the same area for over a decade that are still teaching shit and bore their students that much that they all fall asleep during those lessons... There goes your argument

So all in all, I totally agree with you about that marketing BS :-(

Regards.

 

[maverickronin]

Here are some good explanations.

 

[PenguinMusic]

Hi,

Will give those a look

Thanks...

I'll learn something.

And I'll look for an alternative to the x8.

I am not interested in Bluetooth ability : the x5s 6th (from Aune) already acts as BT transmitter AND as BT Receiver. So that is cool.
I am not interested in a DAC/HP Amp combo as the x7s from Aune already is a HP amp.
And both units working together produce a sound that, well, I like. A LOT !

So, DAC, nothing else... All price goes to DAC, looks and build quality. I must say that the Gustard DAC checks those boxes. But out of my price league unfortunately.

Regards.

 

[PenguinMusic]

Hi guys,

After reading your thoughts in here, I thought maybe I'll reconsider the X8... until it gets reviews from Amir of until I can get someone to perform proper measurement.

So I took the list of devices Amir has tested and that ended up to have good measurement. And I came up with 2 options that checks my boxes.
The Soncoz LA QXD 1 and the Loxjie D30.

They're both available from Amazon with fast shipping (2/3 days) and therefore can be returned easily.

I must say the Loxjie D30 looks like a bargain :
- stellar DAC measurements (unless I am mistaken) ;
- rather nice looks :
- all inputs you can want (USB, optical, coax and Bluetooth) ;
- internal power supply ;
- added bonus of headphone DAC even if not really powerful ;
- AK 4493 DAC Chip ;
- stunningly low price (170€).

The LA QXD 1 has :
- superb looks (really like that minimalistic design) ;
- really good measurements ;
- looks like it has support form the manufacturer (there are updates) ;
- all possible inputs (except Bluetooth which I do not need) ;
- included balanced outputs (even though I dont need them) ;
- decent price (230€ shipped).

Still have to decide, but... the Loxjie seems to be something to consider.

Regards.

 

[Wes]

Hi,

Not sure I really understand all you're writing, but I can say that those kind of claims always makes me somewhat suspicious...

Even for the Aune x8 I'm looking for when they claim that it uses their own FPGA that 6 years to develop.
Me, I don't care about that.
How does it measure/sound ?
That's all I wanna know...

I know of colleagues that are working in the same area for over a decade that are still teaching shit and bore their students that much that they all fall asleep during those lessons... There goes your argument

So all in all, I totally agree with you about that marketing BS :-(

Regards.

Let's try this analogy:

Think of an integrated circuit chip as a very, very complex set of roads with intersections - lot and lots of them.

Now, you want to determine the traffic flow on those roads (cars = electrons). So you can open or close different routes or intersections to control that flow. The network of intersections is like an array of gates.

An ASIC has all the open intersections already determined - we could say "hardwired" - mass production means we can stamp these puppies out for cheap in large quantities because the high cost is in the design, and it is amortized over a lot of chips.

A FPGA let's us send signals down to the intersections and open or close them to control the flow. But each individual chip costs more to make.

So Field Programmable Gate Arrays are suitable for smaller production runs of chips.

 

[ZolaIII]

Let's try this analogy:

Think of an integrated circuit chip as a very, very complex set of roads with intersections - lot and lots of them.

Now, you want to determine the traffic flow on those roads (cars = electrons). So you can open or close different routes or intersections to control that flow. The network of intersections is like an array of gates.

An ASIC has all the open intersections already determined - we could say "hardwired" - mass production means we can stamp these puppies out for cheap in large quantities because the high cost is in the design, and it is amortized over a lot of chips.

A FPGA let's us send signals down to the intersections and open or close them to control the flow. But each individual chip costs more to make.

So Field Programmable Gate Arrays are suitable for smaller production runs of chips.

The part seams you don't understand is what lies in between of hard rugh fixed logic & complex expensive interconnect of programmable gates and ironically it actual name says it all. Everything else in user space at least is pretty much reinventing the hot water.

 

[ZolaIII]

@PenguinMusic Regarding Loxjie while D10 whose great but odd looking fellow the D20 whose a disaster, not really sure why are you so exited about D30 which while good it's not all that cheap and is a real spiritual successor of mentioned odd looking fellow. If you can find D10 & don't mind the looks of it it will probably cost under 100$.

 

[PenguinMusic]

Hi,

As I said, look is important and the D10 definitely does not fit for me.
I might have considered it if I would have used it all alone. But next to x5s and x7s... Nope :-(
Besides, it seems to be difficult to get at the moment.

The D20 was nice, but obviously measured poorly. So poorly a defective device should be suspected :-(

All other things : don't like the look. Or if I like it, the price does not like me :-(

I forgot to mention the SMSL SU-8. Like that one too... and has rather good measures.

But this may not be the right place to discuss this thing any further...

 

[Philbo King]

An interesting side note on FPGAs based on my 27 years in avionics design:
As altitude increases there is a nonlinear increase in neutron single-event upsets (NSEUs), where an energetic particle collides with a bit cell in RAM and flips the bit state of the data in the cell. In FPGAs they use a small local RAM that is loaded on powerup to direct the chip to activate certain digital processing and disable others. If an FPGA RAM cell gets its bit flipped it is, in effect, randomly rewiring the FPGA digital gates on-the-fly. While that might be inconvenient for an audio processor, it is disastrous for a safety-critical flight control function.

There have been a number of methods developed to mitigate this risk, which are beyond the scope of an audio forum thread, but I thought I'd mention it as a point of interest.

And presumably, if any of you use FPGA based audio processors on a SpaceX or Amazon flight, keep in mind it may inexplicably crap out on occasion, only to resume working after a reset or power cycle. LOL

 

[valerianf]

As said Amir, the problem with ASIC development is that it costs a bunch of money, take sometime 1 year between re-spin and it works/does not work without any identified root cause. Sometime you just have to change the fab building in the same foundry to get a completely different end result!
It is black magic.
FPGA allows a much fine control of the design and many more iterations in a short time.

 

[chris719]

An interesting side note on FPGAs based on my 27 years in avionics design:
As altitude increases there is a nonlinear increase in neutron single-event upsets (NSEUs), where an energetic particle collides with a bit cell in RAM and flips the bit state of the data in the cell. In FPGAs they use a small local RAM that is loaded on powerup to direct the chip to activate certain digital processing and disable others. If an FPGA RAM cell gets its bit flipped it is, in effect, randomly rewiring the FPGA digital gates on-the-fly. While that might be inconvenient for an audio processor, it is disastrous for a safety-critical flight control function.

There have been a number of methods developed to mitigate this risk, which are beyond the scope of an audio forum thread, but I thought I'd mention it as a point of interest.

And presumably, if any of you use FPGA based audio processors on a SpaceX or Amazon flight, keep in mind it may inexplicably crap out on occasion, only to resume working after a reset or power cycle. LOL

Yeah, it’s easier to shield than get rad hard components. I’ve designed components in a device with a strong gamma source. We use all off the shelf stuff, including an FPGA, but there’s some lead. Not desirable for space but probably doable for commercial aviation. It becomes a problem for even commercial ASICs like high end CPUs as nodes shrink to “7 nm” and beyond.

 

[Bathrone]

Great thread

No doubt ASICS are the way to go. I will add, there is significant risk reduction in being able to conduct these days a very large amount of testing through simulation test harnesses, which are proven and known to be representative test harnesses pretty much identical to how the mask turns out when fabricated. Many would invest a large portion of their overall testing effort across the systems development lifecycle into test harness simulations. There is still reasons why a new stepping for the mask might be needed.

The subject of bit flipping and environmental radiation causes of that phenomenon. My problem there is unless your in space where shielding mass is a problem, and unless your in space not protected to some extent by earths atmosphere - the specialised chips made for such use cases just plain suck. They are forced into very very old lithography processes and you end up with chips with ancient gate lengths. They are slow, power hungry and hot. Much better for any critical device not entering into space to adapt shielding if needed, and to do other hardening techniques like with bit flips if they do occur, to do so under ECC control schemes to fully manage the event in data critical situations.

Plus, the rate of environmental caused bit flips on the ground in earth is way less than in space. It depends on the gate length of the lithography as to how susceptible a given chip is also.

BTW, I have never understood what Mr Rob Watts at Chord has been banging on about with this hardware choices. One he claims that the FPGA's hes using are cutting edge and not previously available in the market which according to him has historically held back his ability to explore further tap lengths in digital filters and so on. This is false. The FPGA's hes using are cheapish budget models. Also, this makes no sense either in that even if hypothetically what he claims is bizarrely true about hardware, its standard practice in digital processing to do things like symmetrical processing with more than one FPGA. A great example, is purpose built brute force cracking machines for decrypting say AES, where many thousands of FPGA's are sometimes used all at once within a given device. An advanced DAC with say 4 or 8 FPGA's isnt a blocker given the pricing in the high end audio industry

The other thing I have never understood about his claims is his assertion that our most modern, most powerful desktop and server microprocessors can't be used here because they lack the processing needed. He went on in one video to try to assert that 64bit CPUs are too limited in being 64bit. I'd contest that - not everything has to be done in one clock now does it. Massive examples of 1024, 2048 bit processing on 64bit platforms. I realise everyone can make mistakes from time to time but it is indeed pretty dam ignorant to try to suggest a 64bit processor cant deal with bigger data sets. It makes me suspicious of some of his other claims because its hard to reconcile his depth of understanding in specific areas against clearly false and basic things anyone professional should know. I'd certainly never want to suggest its about money and marketing, Id like to think hes better than that.

 

[raif71]

An interesting side note on FPGAs based on my 27 years in avionics design:
As altitude increases there is a nonlinear increase in neutron single-event upsets (NSEUs), where an energetic particle collides with a bit cell in RAM and flips the bit state of the data in the cell. In FPGAs they use a small local RAM that is loaded on powerup to direct the chip to activate certain digital processing and disable others. If an FPGA RAM cell gets its bit flipped it is, in effect, randomly rewiring the FPGA digital gates on-the-fly. While that might be inconvenient for an audio processor, it is disastrous for a safety-critical flight control function.

There have been a number of methods developed to mitigate this risk, which are beyond the scope of an audio forum thread, but I thought I'd mention it as a point of interest.

And presumably, if any of you use FPGA based audio processors on a SpaceX or Amazon flight, keep in mind it may inexplicably crap out on occasion, only to resume working after a reset or power cycle. LOL

I suppose there might a be practical use for NSEUs...maybe a "true" random number generator circuit module ?

 

[Philbo King]

Yes... the elapsed time between nuclear decay events is one of very few truly random events. Using background radiation as a source doesn't work well though; it changes dramatically with your proximity to unstable isotopes.

 

[Tks]

Yes... the elapsed time between nuclear decay events is one of very few truly random events. Using background radiation as a source doesn't work well though; it changes dramatically with your proximity to unstable isotopes.

I don't get it, do error correction fail-safes not exist for such designs, something akin to ECC RAM for example? (I'm talking about your bit flips post, not this RNG generator topic).

 

[muslhead]

Yeah, it’s easier to shield than get rad hard components. I’ve designed components in a device with a strong gamma source. We use all off the shelf stuff, including an FPGA, but there’s some lead. Not desirable for space but probably doable for commercial aviation. It becomes a problem for even commercial ASICs like high end CPUs as nodes shrink to “7 nm” and beyond.

Hmm, news to me
As the past director of highrel space division at a fpga company, we had no problems making or passing NASA space radiation requirements for their satellites. I guess the almost $100m in sales we did annually to THIS MARKET were just a front for what, bicycles?

 

[chris719]

Hmm, news to me
As the past director of highrel space division at a fpga company, we had no problems making or passing NASA space radiation requirements for their satellites. I guess the almost $100m in sales we did annually to THIS MARKET were just a front for what, bicycles?

Maybe you should read better. I mean the garden variety commercial parts. I doubt you’re launching commercial grade parts into space. I’m sure you’re quite able to offer a space qualified menu of parts with greater ability to handle SEUs.

Or are you telling me all your parts were fine in space but the space grade was just to extract more money from government organizations?

Were you making 7nm parts back then? Everyone is more affected as we continue with the process shrinks.

 

[muslhead]

Maybe you should read better. I mean the garden variety commercial parts. I doubt you’re launching commercial grade parts into space. I’m sure you’re quite able to offer a space qualified menu of parts with greater ability to handle SEUs.

Or are you telling me all your parts were fine in space but the space grade was just to extract more money from government organizations?

Were you making 7nm parts back then? Everyone is more affected as we continue with the process shrinks.

Wow, no need to be defensive.
Your lack of knowledge about goverment organizations and the ability to extract more money is quite laughable. Its clear your little feelings were hurt.
Sorry dude. I hope it doesnt ruin your day.

Just so you dont have to have more angst, no it was not 7mm back then and you are absolutely correct regarding process shrinks. But our process was not the same and we were able to scale the rad hardness with our shrinks. Still the same today? Doubt it as you can only do so much with smaller gate sizes

 

[chris719]

Wow, no need to be defensive.
Your lack of knowledge about goverment organizations and the ability to extract more money is quite laughable. Its clear your little feelings were hurt.
Sorry dude. I hope it doesnt ruin your day.

Just so you dont have to have more angst, no it was not 7mm back then and you are absolutely correct regarding process shrinks. But our process was not the same and we were able to scale the rad hardness with our shrinks. Still the same today? Doubt it as you can only do so much with smaller gate sizes

I’m not being defensive, I said nothing that was incorrect in my first post, but in comes big D swinging “I worked at xxx”. Great, I’m sure you have an impressive CV. I didn’t say FPGAs were unsuitable for space, just *off-the-shelf” aka commercial grade. Which is true. I never meant to imply that there are no FPGAs in space. That would be stupid.