Slashdot Mirror
A Workstation for Sensitive Experiments?
etrgQUARK asks: "I am in charge of infrared spectrometers at our research center. One of the setups is used to measure the orientation of monolayers at the air/water interface, i.e. the signal we have to detect is very weak and noisy. We already have a great setup with quality components and electronics, except for one piece of hardware: the computer hooked up to acquire the data. How important is the computer in noise-sensitive data acquisition? What are your experiences? Do you have any suggestions on a workstation suitable for such tasks or is it a waste of money to use anything but the average computer system? Unfortunately, the software used is Windows-only."
You should look at expensive, custom data logging equipment for aquiring the information you need. When you have that you can shield it properly. The manufacturer's specs will show the sensitivity/noise/etc and you can select one that can actually record your signals with reasonable resolution.
Don't run a PC anywhere near it if it's as sensitive as you say. PC's generate a lot of noise and they'll interfere with practically any sensitive measurements... take for example your TV. The TV isn't particularly sensitive but your PC can create noise on some of the channels.
Just hook the data logger up to a PC after the experiment is complete.
I drink to make other people interesting!
Uh, by the time it gets to your PC it should be digital. So "noise" is not a worry unless you're planning on doing the A-to-D on the PC, in which case you need a psychiatrist, not
--MarkusQ
Shouldn't this depend entirely on where the A-to-D conversion happens and how much EM interference the computer produces?
As for the Windows dig, I can't see how Linux would result in less interference.
What sort of signal's are you recording exactly? Will they survive a long cable jaunt?
Move box away from experiment. Problem solved.
Are you worried about noise once the signals reach the machine internals (if they do)? If so, why would there be a internal data logger that can't log data correctly even on the market?
With a 16 bits ADC you might get 14 bits effective resolution at the 10 V input range. The last 2 bits are often mostly noise even if you shortcut the input. Always use differential mode instead of single-ended such that noise from ground loops is eliminated. (It means it does an analog substraction of the signals on two inputs, rather than compare them to the common ground that may be noisy)
If your noise requirements are much higher, then the best thing to do is to amplify the signal you're looking for before it goes into the ADC board. Use lock-in techniques if you can. For example if you're trying to see variations of 0.001 V on top of a 5 V signal, find a way to modulate the 0.001 V signal (e.g.chop the light source at 1000 Hz) and use a lock-in amplifier to measure the oscillating 0.001 V component and amplify it to some value that is easier to send to an ADC.
Avantslash: low-bandwidth mobile slashdot.
And of course, not entirely dependent on the "sensitivity experimentation" factor, but still: make sure to backup the results as much as possible. Try to have at least one external store of the data. After all, that could be one of the weakest links in your long chain of experimentation.
Would this be an increase in noise?
The ______ Agenda
it's hard to really say. i imagine that you have a daq board with high precision A2D converters and sufficient buffering that so long as you can sustain the throughput you're not going to lose any data. (if not, you'd better fix this bit first)
PC side, I'd suggest attention to the IO and memory subsystems if your data is arriving at a sufficiently high rate (of course, this requires your acquisition software to be written to take advantage of the hardware). that said, my suggestion would be: at least 1gb ram (paired modules), separate raid volume for the incoming data (1+0). the rest is dependant on your analysis needs — does your acquisition app do analysis too? would it benefit from multiple cores?
you'll want a vendor with top notch support (so be willing to shell out the extra $whatever for the top end support contract). make sure you explain the importance of your system to them. also, make sure you're going to get support from the software vendor for your acquisition and analysis applications.
other than that, you're pretty much on your own i'm afraid -- you know your application better than i do.
good luck!
You're seriously posing this question to a forum full of GNAA posts, The Giver links and "frost pist!" comments?!
Even though you're talking about something scientific, i'd like to point out that there is a very high demand for low-noise computers in the home-recording scene. (Low noise also equals low hum/shaking of the machine)
Phil Rees computers are built to be extrimely quiet (and frankly i think your mouse makes more noise that one of these computers. www.philrees.co.uk
I don't know if this helps.. probably better than most of these troll answers.
You didn't mention the kinds of noise you're interested in keep out, or what kind of data your sensors return. I will therefore assume, "all" noise and many kinds of sensors.
The best approach to keeping something as electrically noisy as a PC from spoiling your results is to put it in another room and connect it only with radio or light. This also addresses the sound issue.
If the bandwidth on your sensors is low enough for RS-232 serial data, then you're in luck, dozens of manufacturers sell simple in-line isolators like this.
If the cable it's running through is still picking up too much noise then using fiber converters on both ends (like this) will let you bridge the gap with glass which is pleasantly resistant to electrical noise.
If you're currently capturing your data with a bunch of low sample rate A-D converters, and have a large wad of sensor lines going to the attachment-pod on the PCI card, now would be the time to get your university's micro controller enthusiasts to create a sensor polling device. On something like the Atmel 8515.
Such a sensor converter will run for days on batteries, and produces very predictable low amplitude noise that is easy to isolate. It comes with an 8 channel 10-bit A-D, and the best part is they are available in an easy to breadboard 40-pin dip for $4.58 through digikey (here search on ATMEGA32-16PJ-ND).
If the bandwidth of your sensors is too great for RS-232, the same tricks can be played with RS-485 transceivers which will do speeds into the megabits, and are available as cheap dip packages.
For speeds beyond the few megabits realm a "custom data gathering CF card" and a Wifi capable IPAQ running Familiar springs to mind as a good starting point. But here we're getting into the question of "What exactly is it you're doing on the Windows box?"
--Art
PCs and long analog cable runs can definitely be a negative, especially if your signals are low level.
My experiences with this sort of stuff is that you want to move the D/A converters as close to the experiment as possible and to use good instrumentation grade wiring with twisted pairs individually shielded plus a drain wire. If also sounds like your setup may be very sensitive to mechanical vibrations - if your noise source is mechanical nothing electronic will really fix the problem. You can filter stuff in the digital domain but you lose frequency response when you do.
I've had very good luck with Analog Devices D/A stuff in the past; not particularly expensive and pretty good quality modules that you use in a distributed fashion to get into the digital domain as quickly as possible.
Please give us the name of your boss, so we can let him know that you have no idea how to do the job you're "in charge of". Also an address to which I can send my resume.
You should try talking to the people at National Instruments. You can have the best electronics setup in the world, but if you don't have an appropriate way of digitizing the data, your results will suffer. If all of your digitization is being done elsewhere and you don't have a ton of data to save, then you won't need a very good computer, but you also won't be able to use the computer as an active part of the experiment.
You can digitize the data just fine at the computer, just amplify and buffer it beforehand. Don't use a sound card, you will have difficulty with DC signals or offsets, and regular DAQ cards are available for similar prices. Gathering data directly is always preferable to using some sort of digital middle-man such as GPIB, because it will allow you to respond to the data much more quickly. A computer can be used to mimic and improve the effects of many other devices. Most lock-in amplifiers these days are digital anyway, why not have complete control of it? If you're good, you can get a whole lot of funtionality out of what is really a very cheap piece of hardware.
My qualifications? Single cell exocytotic measurements using 5 micron carbon fiber microelectrodes.
And for the love of god do not attach the computer to the internet ever. Collect data and if you have to burn it to a CD and sneaker net it to machines that are.
I've hit Karma 50 and gotten a Score:5, Troll... I win!
One of the easiest thing that a noise-analysis hobbyist can do is configure the PC BIOS.
1. Spread-Spectrum is a MUST enable.
aka Clock Spread Spectrum
aka Spread Spectrum Modulation
Other EMI-reduction methods are:
2. Older and slower PC have better noise level (may conflict with DAQ adapter requirements)
3. Underclocking as much as possible on higher Ghz CPU. I'd prefer older and slower CPUs.
4. A GOOD Metal Case. Aluminum isn't worth crap (Slashdot also had a tin-foil hat story about aluminum's ineffectual shielding for USA passport's RFI). Added Copper then Iron/Steel foils around the case are best. Screened wires of very small apeture.
5. Run shielded cables for IDE (80-wire 40-pin preferably); SCSI
6. Turn off Ethernet adapter, USB 2.0, firewire.
7. UPS is a god-send and king of AC spike filtering, use it with your PC.
8. Monitor are the worst offender in EMI emanation. Use LCD panel and lowest refresh rate.
Every little bit goes a LONG way to a nice quiet electronic lab.
I design and deploy custom data logging solutions, signal conditioners, etc.
Computer at the end of the chain should not be a consideration whatsoever. The system under monitoring should be completely seperate; by the time the computer is involved, it should be recieving a conditioned, pre-amplified, or digital data stream over a galvanically/optoisolated connection.
If the introduction of a computer device causes a problem, there are other issues to consider.
..don't panic
Yeah, you weren't the only one thinking the radio suggestion was deleriously misinformed. Any networking here should be done over shielded twisted pair, honestly I'd suggest token ring. Fiber transceivers might make just as much EMI as anything else. If a 33.6k data rate is acceptable, modems will happily work over cabling with serious chokes on it. (And remember: Only ground one end of the shield.)
:)
Another poster mentioned CRTs as the major EM noise source. Seconded! Video cables are nasty too. If you want to build the whole PC into a metal box, I've seen shielded windows available. Probably expensive for larger sizes though.
Depending on what form your data acquisition takes (Hint: USB!), a laptop might be the most appropriate form factor here. I own a ruggedized laptop with a metal case, and adding internal wireless cards is nearly impossible, because the RF has a hard time leaking out! Also, laptops are fed with DC, which you can filter more easily than the AC input of a desktop. Find one where the case is the heatsink, so there's no fan, for reliability and noise reasons.
An earlier poster mentioned mechanical noise. A solid state disk might be a good way to eliminate mechanical rattling and clattering in the computer itself. A number of outfits offer adapters that'll let you mount CF cards in place of laptop or desktop hard drives, and CF in the gigabyte range is surprisingly affordable now.
Don't forget about software like rdesktop and VNC, if you need to control the DAQ software from afar without making footsteps in the lab. You could actually run a headless computer for acquisition, and do all the control remotely.
If you could provide some more detail about the experiments, the data acquisition hardware, and the problems that others in the field have run into, I'm sure we could suggest more ideas. Ask Slashdot can be a full-duplex conversation.
2. Test lead shields grounded at ONE point and ONE point ONLY.
3. DO NOT use serial cables (RS-232, USB, etc.) between test equipment and computer.
4. Spectrometer? Sensitive to noise from computer? Make sure all equipment covers are in place and fastened!
5. Fluorescent Lights can be a problem to investigate and identify.
6. One electric power phase. I once spent a couple frustrating hours to find side by side outlets were on different phases. Quality power strips are a must!!
The only thing new in this world is the history that you don't know.[Harry Truman]
Often, there are significant sources of noise within the reception or data capture device (such as the ADC). In these cases, it can be beneficial to amplify the input signal to that system.
For example, suppose those 2 bits of noise are from the ADC unit itself. In this case, amplifying the input signal prior to reaching the ADC is beneficial. This especially holds true if the ADC is inside a computer system.
Another good example is placement of a preamplifier before vs. after a long coaxial cable run. It can be shown mathematically that the total noise figure of the amplifier + coax system (SNRin/SNRout) will be lower if the amplifier is placed before the lossy cable rather than after it. (Lower is better.)
retrorocket.o not found, launch anyway?
I use similar equipment in my lab. CPU speed hasnt been much of an issue, but never never NEVER hook it up to the internet or a network.
preferablly connected via a low speed optical link (say rs-232 run over simple plastic fibre stuff nothing fancy needed)
note: i'm known as plugwash most places but i screwd up registering that here somehow in the past and now can't register
I had to deal with this many years ago. The two biggest sources
of interference were the CRT display and the switching power supply. That was pre-LCD so I had to live with the CRT but replacing the power supply with a linear regulator helped reduce noice significantly.