ARM, Single-Chip PC, or other architectures for audio stuff?

The Intel Galileo board has come up in this newsgroup but I don't remember if it was in this thread, so just in case, it may be of interest:

formatting link

It uses the Quark SOC which is a wimpier (and lower powered) x86 than an Atom.

Reply to
Paul Rubin
Loading thread data ...

The system must be absoluetly power-loss proof. The power plug could be pulled at any given moment and there will not be any accessible user interface. Not even one lone button. There may be some on the box for the occasional configuration event but generally the box is out of reach of any person.

Actually this one will just plow along at a continuous horsepower, very constant load.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

That looks very interesting. Clicking on technical details pulls a blank though :-)

In case anyone else needs it, a working link is here:

formatting link

The only thing I absolutely do not want to deal with is the Arduino IDE. But I assume one doesn't have to. The other fly in the opintment is that the SOC is a BGA. That can spell trouble for gear that can be handled roughly or be dropoed to the floor on occasion. That's the beauty of ARM chips, they come in gull-wing flat-packs.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

Did you look for PC104/PCI104(Express) stacks ? These should fit the size requirements.

In an other post, you indicated that you are going to use Win XP, but would Win XP Embedded be sufficient ? At least you can get away of much of the useless CPU load and hence electric power consumption, by turning off much of the UI eye candy, such as smooth window updates.

What kind of load profile do you want from the processor ? As much computations that you can get without overheating or a constant number of CPU cycles each second ?

In the latter case the average power (or total energy consumption) is nearly equal, regardless you run the processor for 1 ms at full clock every 10 ms or run it for 5 ms every 10 ms at Fmax/5.

In some cases running a powerful processor at Fmax/5 allows running it at a slightly lower operating voltage. Since the power consumption is typically proportional to the square of the operating voltage, some real gains can be achieved in power consumption.

In addition, if high speed processing needs to be done rarely (say once a second), one should also study the various processor sleep modes and their wakeup times.

Reply to
upsidedown

Do all these atom based things come without a BGA? I would be surprised if so (last time I looked at an Intel processor was around 586, I checked it was still the same architectural mess I knew from earlier checks).

I think I already suggested the mpc5200b (or similar, the 5125 and 5121 are also options), you will get all the horsepowers you need at 1W CPU power - total system power will be more like 2-3W, DDRAM etc., but it is a nearly 800 MIPS processor, likely more efficient than a 1GHz atom, with all the peripherals etc., there will be little except it, a flash and some DDRAM (I typically put 2 chips, allows me either 64M or 128M) to take in your ADC data and to process it. I could probably be some help if you go this way.

Dimiter

------------------------------------------------------ Dimiter Popoff, TGI

formatting link

------------------------------------------------------

formatting link

Reply to
dp

Right, the Galileo runs Linux.

I thought one normally deals with this by potting the circuit.

Reply to
Paul Rubin

That's good, it's just that the frequent mention of Arduino "shields" in there got me spooked. I am a Windows guy but Linux won't be a problem in this case.

You can't really take the stress on the solder balls away by potting. Unless it's all encapsulated in glass maybe :-)

Worst case potting makes it worse because of differences in thermal expansion versus the FR4 and copper.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

It is certainly an option but the MPC only comes in BGA, if I read correctly.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

So far I found them power-hungry and there were BGA on them which I am trying to avoid on this system.

I don't know Win XP Embedded but most likely the answer is yes. We don't need any eye candy. This system will have no user interface when operated normally. Seriously, the main power plug is the only "user interface", on or off.

Essentially two 1024-point complex FFTs running at a 5msec clip, plus another 30% of that as overhead math horsepower for pre- and post-processing. No fans, not heat, other than a watt or two.

This is going to be a very even workload running all the time, until someone (literally) pulls the plug.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

Or die bonding directly.

It's hard to build any large memory system (1G or more, for Win or Linux) with less than 200 pins. Large non-bga packages also have other problems.

Reply to
edward.ming.lee

Yes, this is something you will have to deal with somehow, I expect the cost you will pay in dealing with that will be small compared to the compromises you will have to make falling back to only what you get in qfp etc.

Dimiter

------------------------------------------------------ Dimiter Popoff, TGI

formatting link

------------------------------------------------------

formatting link

Reply to
dp

For applications like this it doesn't need to be large. A 100-QFP would be fine. That's one reason I'll be looking at ARM because they come in such packages.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

The cost for field failure is painful in this case. It goes into a field where folks have a very low tolerance for things going kablouie because this brings everything to a screeching halt. So it pays to thoroughly scope out and see if there is a decent non-BGA that can do the job with sufficient margin. Looking at the ARM-world it seems there is.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

I meant the cost for you to find a solution with BGA so your product is uncompromised. BTW I am not sure at all a large QFP will do mechanically better than the BGA, I have seen pads of these get torn away while I have not had a BGA fail me mechanically. If you are worried about board bending you can address the bending itself somehow (2mm board, I don't know).Anyway, that's work you will have to do. The 5200B is on a 0.5mm-ish board which will likely bend together with the PCB, I am sure I have subjected my prototype boards to more abuse yours will face when thrown around in the field and I have had no such failure (I do have a via on the PCB in the centre of each BGA pad, this might be helping).

But going MPC5200B will likely save you a lot of the horror I constantly hear from people using half-baked ARM parts (are there any others? :-) ). I am not saying the 5200B is perfect, I spent 3-4 months myself until I had all the BIOS stuff ready to boot DPS and there were bugs on the silicon, but no real killers. The worst one was that its smart DMA - which is the best DMA I ever saw - did not do misaligned transfers; it cost me a while until I did all that using DMA only (read one place, copy elsewhere - SDMA only) for the disk I/O, add to that all the physical page lists... It was a thing to do :-). Well there was one more almost killer, the ATA interface would hang once every few minutes of steady transfer, but this was recoverable and quickly detectable so the fix is unnoticeable.

Dimiter

------------------------------------------------------ Dimiter Popoff, TGI

formatting link

------------------------------------------------------

formatting link

Reply to
dp

I've just seen too many BGA failures over the years. One problem is that you cannot inspect the solder joints in any reasonable way. With QFP you can. I never had a properly soldered QFP fail on me.

I know a few guys over in Europe who have successfully implemented difficult projects on ARMs. Seems they are quite popular there. Then there are whole smart phones built around them and ... well, they work.

It sure would save time if I had someone local who can support a certain architecture. That might swing us to whatever that architecture or processor is (preferably non-BGA though). Nothing beats experience in that area.

--
Regards, Joerg 

http://www.analogconsultants.com/
Reply to
Joerg

The via I put in the centre of each BGA pad solves this - you just see after reflow most of the holes being filled, some (smaller balls) not but if you see all around the BGA holes which got full you know it is OK.

The disadvantage of that is that you _have to_ bake the BGA with some decent flux once balls up prior to mounting; every once in a while a ball or two per BGA come from the manufacturer "coldly" attached which is not a problem on pads with no holes and flux but they do flow through holes (all of the ball is gone). The baking balls up solves the issue completely, just use decent flux (I usually use a "Loethonig" I get at Buerklin :-) ).

Dimiter

------------------------------------------------------ Dimiter Popoff, TGI

formatting link

------------------------------------------------------

formatting link

Reply to
dp

Do you have room for a motorcycle battery and a 12V-5VDC switcher? Just trickle charge the battery off the wall...

And there's this:

formatting link

A standalone lithium-ion 9V source. No clue if it can be charged while operating.

--
Les Cargill
Reply to
Les Cargill

Unless you have severe mass (aviation/space) or volume limitations, have you considered using double or even triple (voting) redundancy ?

By using redundancy, more or less COTS components/systems can be used and getting a higher system reliability compared to a single high reliability (and expensive) component.

If the cost of one day loss of production is high enough, the customer will gladly accept a redundant system, in fact in most cases demanding a redundant system.

Reply to
upsidedown

The power consumption is directly proportional to the number of CPU cycles you are going to burn, thus minimizing the number of CPU cycles should be the obvious. In a headless system (no user interface) I do not understand why you need such a high performance processor, unless you do some serious DSP at high sample rates, as you apparently are doing.

A few years ago I evaluated some Windows compatible x86 processors for low power/high temperature operations. At least the VIA processors could be quite easily underclocked, dropping the power consumption quite low. Unfortunately, if you need to perform a constant number of CPU instructions each second, this does not help much, the total energy consumption will be the same, unless you can drop the operating voltage.

Regarding BGA unreliability, check my other comment about building a redundant system.

If the "user interface" is just a VGA connector for maintenance, you should study the Windows XYZ Embedded variations.

Thus you have a well defined base load, underclock the CPU, until the base load consumes about 60-80 % of the available CPU cycles. Drop the CPU core voltage to the lowest that still runs at that CPU frequency reliably, in order to get the lowest power consumption.

Reply to
upsidedown

You are sampling 1024 times in 5 msec, right? That's over 200KHz ultra-sonic. How many clips do you need to buffer up? That would determine whether you need microcontroller (Arm) or microprocessor (Atom).

If you are doing ultra-sonics imaging, you need Atom.

Atom can be fanless.

Atom runs at 1.6GHz with less than 2W. With sleep or suspend, it can be much lower. Not sure if it can be underclocked.

Reply to
edward.ming.lee

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.