Best way to getting a high data rate (300Mbps) from multiple ADCs to memory

Hi,

I am wanting to simultaneously sample 100 channels of microphone data with
a resolution of 24 bit, at 128 kSPS (kilo samples per second) per channel.
I then want to save the data to file for later analysis. I assume that 
recordings will be made for a few seconds at a time.

This will require a data throughput rate of approximately 300 Mbits per
second, ie. (100 Channels)*(128 kSPS/channel)*(24 bit). 

I am not sure what is the best way to get this data from the ADCs to
memory. Is one or more DSP or FPGA the best way to go and if so which ones.
I have been looking at the TMS320C64x DSPs but perhaps there is something
better/easier.  Also what is the best interface to get the data from any
such device to the memory. Also what type of memory would be best. I am new
to     all this so any advice you could give would be gratefully
appreciated. Thanks     

mlphy wrote:
> Hi,
> 
> I am wanting to simultaneously sample 100 channels of microphone data with
> a resolution of 24 bit, at 128 kSPS (kilo samples per second) per channel.
> I then want to save the data to file for later analysis. I assume that 
> recordings will be made for a few seconds at a time.
> 
> This will require a data throughput rate of approximately 300 Mbits per
> second, ie. (100 Channels)*(128 kSPS/channel)*(24 bit). 
> 
> I am not sure what is the best way to get this data from the ADCs to
> memory. Is one or more DSP or FPGA the best way to go and if so which ones.
> I have been looking at the TMS320C64x DSPs but perhaps there is something
> better/easier.  Also what is the best interface to get the data from any
> such device to the memory. Also what type of memory would be best. I am new
> to     all this so any advice you could give would be gratefully
> appreciated. Thanks     
> 
What environment are you working in?  Are you trying to get this data 
into a PC, or will it stay on some sort of embedded board?  Is space a 
consideration?  Are you building one system or 1000?  Do the 100 
channels have to be synchronized (I'm assuming you're doing some sort of 
localization and the answer is 'yes').

300Mbits per second for a few seconds isn't theoretically challenging -- 
DRAM is cheap, and spinning ten seconds worth of data into DRAM would 
only require 400MB of DRAM.  The hard part is acquiring the data from 
100 channels and handling it in a manner that doesn't require more wires 
than exists in all of creation.

One thing that jumps to mind immediately is to build or acquire 100 
identical small data acquisition systems, each capable of collecting 
data from one microphone, buffering a few seconds worth, and sending the 
information to the PC via Ethernet.  Synchronization would be an issue, 
but can (I assume) be handled.

Have you investigated the possibility of finding PCI data acquisition 
that does what you want or part of what you want, and stuffing a bunch 
of them into a PC?  Or a number of PCs, with external synchronization? 
Or a PC with a PCI extension bus (I _think_ you can get those)?

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" gives you just what it says.
See details at http://www.wescottdesign.com/actfes/actfes.html

Hi, thanks for your reply.

>What environment are you working in?  Are you trying to get this data 
>into a PC, or will it stay on some sort of embedded board?  

I want to get the data to a PC. A short recording will be made and then
the  data will be transfer to the PC.

>Is space a consideration?  
>Are you building one system or 1000?

At this stage, space is not a consideration. Compactness is not an issue.
I will be building only 1 to 3 units at this stage.
  
>Do the 100 channels have to be synchronized (I'm assuming you're doing
>some sort of localization and the answer is 'yes'). 
>One thing that jumps to mind immediately is to build or acquire 100 
>identical small data acquisition systems, each capable of collecting 
>data from one microphone, buffering a few seconds worth, and sending the

>information to the PC via Ethernet.  Synchronization would be an issue, 
>but can (I assume) be handled.
>

I am considering using an ADC which allows synchronize all channels. The
output of theese ADCs is either SPI or frame sync format. 

>Have you investigated the possibility of finding PCI data acquisition 
>that does what you want or part of what you want, and stuffing a bunch 
>of them into a PC?  Or a number of PCs, with external synchronization? 
>Or a PC with a PCI extension bus (I _think_ you can get those)?

Yes I have looked into getting a 100 channel simultaneous sampling ADC
system but everything I have seen is out of my budget. I had wondered if
one could get some sort of pci data input board to take the data from the
ADCs to the PC, which would be less expensive.  Thanks

>Hi, thanks for your reply.
>
>>What environment are you working in?  Are you trying to get this data 
>>into a PC, or will it stay on some sort of embedded board?  
>
>I want to get the data to a PC. A short recording will be made and then
>the  data will be transfer to the PC.
>
>>Is space a consideration?  
>>Are you building one system or 1000?
>
>At this stage, space is not a consideration. Compactness is not an
issue.
>I will be building only 1 to 3 units at this stage.
>  
>>Do the 100 channels have to be synchronized (I'm assuming you're doing
>>some sort of localization and the answer is 'yes'). 
>>One thing that jumps to mind immediately is to build or acquire 100 
>>identical small data acquisition systems, each capable of collecting 
>>data from one microphone, buffering a few seconds worth, and sending
the
>
>>information to the PC via Ethernet.  Synchronization would be an issue,

>>but can (I assume) be handled.
>>
>
>I am considering using an ADC which allows synchronize all channels. The
>output of theese ADCs is either SPI or frame sync format. 
>
>>Have you investigated the possibility of finding PCI data acquisition 
>>that does what you want or part of what you want, and stuffing a bunch 
>>of them into a PC?  Or a number of PCs, with external synchronization? 
>>Or a PC with a PCI extension bus (I _think_ you can get those)?
>
>Yes I have looked into getting a 100 channel simultaneous sampling ADC
>system but everything I have seen is out of my budget. I had wondered if
>one could get some sort of pci data input board to take the data from
the
>ADCs to the PC, which would be less expensive.  Thanks
>
>

My suggestion is to go with ADC's that support TDM Mode.  Using this
method you can transfer 8 or 16 channels in only three wires.  I'd then run
this to a couple DSP's.  Your main problem here is going to be peripheral
support.  TI's McASP's work nicely but at some point you'll be limited to
something like 16 channels per McASP port and if you're part only has 3
McASP ports... you're looking at 3 DSP's.  

My personal preference for DSP's nowadays is the Analog Devices Blackfin
line.  I've found TI's parts to be very challenging when setting up I/O
ports (near 100 registers to setup!!)  while Blackfins tend to be more
simplistic.

Then strap on a big old SDRAM and just use DMA to stream from input port
to SDRAM buffers.  Then you'll probably want some kind of USB driver on one
DSP (the one processing the least amount of channels) to transfer data to
pc.


All-in-All it's a pretty big project and will take some time to get it all
running correctly.

Good Luck!
-Jeff

On Tue, 20 May 2008 19:05:00 -0500, mlphy wrote:

> Hi, thanks for your reply.
> 
>>What environment are you working in?  Are you trying to get this data
>>into a PC, or will it stay on some sort of embedded board?
> 
> I want to get the data to a PC. A short recording will be made and then
> the  data will be transfer to the PC.
> 
>>Is space a consideration?
>>Are you building one system or 1000?
> 
> At this stage, space is not a consideration. Compactness is not an
> issue. I will be building only 1 to 3 units at this stage.
>   
>>Do the 100 channels have to be synchronized (I'm assuming you're doing
>>some sort of localization and the answer is 'yes'). One thing that jumps
>>to mind immediately is to build or acquire 100 identical small data
>>acquisition systems, each capable of collecting data from one
>>microphone, buffering a few seconds worth, and sending the
> 
>>information to the PC via Ethernet.  Synchronization would be an issue,
>>but can (I assume) be handled.
>>
>>
> I am considering using an ADC which allows synchronize all channels. The
> output of theese ADCs is either SPI or frame sync format.
> 
>>Have you investigated the possibility of finding PCI data acquisition
>>that does what you want or part of what you want, and stuffing a bunch
>>of them into a PC?  Or a number of PCs, with external synchronization?
>>Or a PC with a PCI extension bus (I _think_ you can get those)?
> 
> Yes I have looked into getting a 100 channel simultaneous sampling ADC
> system but everything I have seen is out of my budget. I had wondered if
> one could get some sort of pci data input board to take the data from
> the ADCs to the PC, which would be less expensive.  Thanks

What are your abilities, what is your budget, and what is your time 
worth?  With much messing around you may be able to get a bunch of cast-
off PCs, synchronize them with Ethernet, and use data acquisition cards.  
This could be done with software skill but not much hardware skill.

Alternately, with much messing around and some (relatively) inexpensive 
eval boards, you could do the same thing with DSP chips or perhaps as few 
as one big FPGA.

If you were budgeting this like a business, and valuing your time, it's 
going to cost $$$ no matter how you slice it.

-- 
Tim Wescott
Control systems and communications consulting
http://www.wescottdesign.com

Need to learn how to apply control theory in your embedded system?
"Applied Control Theory for Embedded Systems" by Tim Wescott
Elsevier/Newnes, http://www.wescottdesign.com/actfes/actfes.html

Hi, thanks for your help and advice. Yes it does seem like it is going to
be a bit of a job. I may get funding to buy a ready made unit. However, it
is best, I think, to assume I will not be getting any funding and carry on
making one myself. 

Which do you think would be easier/best to use, DSPs or FPGAs. 

I will look into Analog Devices DSPs. In relation to ti's DSPs, I
understand that the data throughput rate of their McBSP ports are related
to the DSP's cpu speed. In this case would it be better to get one or more
very fast one (1 GHz) or get multiple less fast ones (200MHz).

Thanks    

mlphy wrote:

> Yes I have looked into getting a 100 channel simultaneous sampling ADC
> system but everything I have seen is out of my budget. I had wondered if
> one could get some sort of pci data input board to take the data from the
> ADCs to the PC, which would be less expensive.  Thanks

Did you look for a typical pro-audio solution? This could be cheaper than an
indutsry type ADC system. Maybe you can find something here: 

http://www.rme-audio.de/en_index.php

Usually these boards can be stacked up to multiply the channel number.

bye
Andreas
-- 
Andreas H�nnebeck | email: acmh@gmx.de
----- privat ---- | www  : http://www.huennebeck-online.de
Fax/Anrufbeantworter: 0721/151-284301
GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc
PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc

>mlphy wrote:
>
>> Yes I have looked into getting a 100 channel simultaneous sampling ADC
>> system but everything I have seen is out of my budget. I had wondered
if
>> one could get some sort of pci data input board to take the data from
the
>> ADCs to the PC, which would be less expensive.  Thanks
>
>Did you look for a typical pro-audio solution? This could be cheaper than
an
>indutsry type ADC system. Maybe you can find something here: 
>
>http://www.rme-audio.de/en_index.php
>
>Usually these boards can be stacked up to multiply the channel number.
>
>bye
>Andreas
>-- 
>Andreas H�nnebeck | email: acmh@gmx.de
>----- privat ---- | www  : http://www.huennebeck-online.de
>Fax/Anrufbeantworter: 0721/151-284301
>GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc
>PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc
>
>
Thanks, thats a good idea. I had not thought of that.

>Hi, thanks for your help and advice. Yes it does seem like it is going to
>be a bit of a job. I may get funding to buy a ready made unit. However,
it
>is best, I think, to assume I will not be getting any funding and carry
on
>making one myself. 
>
>Which do you think would be easier/best to use, DSPs or FPGAs. 
>
>I will look into Analog Devices DSPs. In relation to ti's DSPs, I
>understand that the data throughput rate of their McBSP ports are
related
>to the DSP's cpu speed. In this case would it be better to get one or
more
>very fast one (1 GHz) or get multiple less fast ones (200MHz).
>
>Thanks    
>

Personally, I feel that DSPs are the better choice, but I have been using
them for a while now so I am biased.

McBSP ports still have some channel count limitations.  For example they
can sometimes only have two data pins in.  Most TDM ADC's will top out at a
16CH Data Stream.  So that means that you'll have at most 32 Channels per
McBSP and if you're part only left with 64 Channels per DSP.

Plus you may want to look into parts with McASP ports instead of McBSP. 
They are more multi-channel friendly. McBSP's are more tuned in to stereo
sources (I2S format).

If you can find a 1GHz part that has 4 McASP-like ports, this would likely
be the better solution.  The complication factor increases exponentially as
you add multiple IC's.  Look at it this way:

   Complication = e ^ (number of processors)

Just my 2 cents.

-Jeff

Thanks for that advice. I'll look into that. 

I am wondering if it is possible to save data directly from a DSP to a
hard drive? Thanks