-
Welcome back Guest! Did you know you can mentor other members here at H-M? If not, please check out our Relaunch of Hobby Machinist Mentoring Program!
Needing more than a spark test?
- Thread starter graham-xrf
- Start date
Sure Rob - I found that also.
The Pi-4 can get along OK well as it is , but given that this one has the kind of power you find in a general desktop computer, it deserves some sort of heatsink, or it will keep throttle back to 600MHz if it encounters intense computing. Running YouTube + browsing at the same time as compiling does it. The CPU temperature monitor gadget shows 60C to 75C.
Passive heatsinks can help lots - to about 55C, but I found that even the smallest 30mm fan blowing down on the smallest 50cents little set of fins dramatically reduces the CPU temperature - to around 38C to about 48C.
One of those is on the Pi3 Model B, and it is pretty quiet near silent.
The Pi4-B, with the bigger fan is just silent hard to hear, and runs at around 32C. That Pi4 is overclocked to 2GHz. One could take it further, but I kept it to something safe. I have yet to stress it hard.
Normal running is to idle at 600MHz. As soon as a task gets going, it flips to 2GHz. The Raspian desktop is fast and lightweight, but I installed the Mate desktop from the repository. One can use Ubuntu "mate", but I preferred the Debian repository "Buster" that comes with the Raspian. It's a wasteful extravagance, but just way too easy to leave it on Internet Radio, or YouTube Dr Saxlove smooth, and let the Bluetooth speakers have their way.
The Pi-4 can get along OK well as it is , but given that this one has the kind of power you find in a general desktop computer, it deserves some sort of heatsink, or it will keep throttle back to 600MHz if it encounters intense computing. Running YouTube + browsing at the same time as compiling does it. The CPU temperature monitor gadget shows 60C to 75C.
Passive heatsinks can help lots - to about 55C, but I found that even the smallest 30mm fan blowing down on the smallest 50cents little set of fins dramatically reduces the CPU temperature - to around 38C to about 48C.
One of those is on the Pi3 Model B, and it is pretty quiet near silent.
The Pi4-B, with the bigger fan is just silent hard to hear, and runs at around 32C. That Pi4 is overclocked to 2GHz. One could take it further, but I kept it to something safe. I have yet to stress it hard.
Normal running is to idle at 600MHz. As soon as a task gets going, it flips to 2GHz. The Raspian desktop is fast and lightweight, but I installed the Mate desktop from the repository. One can use Ubuntu "mate", but I preferred the Debian repository "Buster" that comes with the Raspian. It's a wasteful extravagance, but just way too easy to leave it on Internet Radio, or YouTube Dr Saxlove smooth, and let the Bluetooth speakers have their way.
If anyone is considering building up their own ADC for this project, one important parameter (other than sampling rate) you want to pay attention to is the ENOB or SNR of the ADC. Most ADCs don't actually produce data that is as accurate as you might think. For instance, high speed ADCs used for DSOs may be labeled as "8 bit converters" but their Effective Number of Bits (ENOB) may only be 6 bits at their maximum sampling rate. ADCs designed for audio purposes often are specified in terms of their SNR, signal-to-noise ratio. A perfect ADC's SNR can be calculated using this equation: SNR = 6N + 1.76, where N is the word length in bits. For example a perfect 16 bit converter would provide a 97.8dB SNR, but you are pretty unlikely to find a 16 bitter that can actually achieve that. For instance, the SGTL5000 that's used for the Teensy audio board is spec'd to 90 dB. While this may not seem like that big a deviation from 97.8 dB, keep in mind that 20log(2) = 6, so the chip is "only" achieving less than HALF the theoretically-possible SNR (it actually is closer to 2.5 times worse).
This is why I'm looking at the PCM4220 which boasts a 123dB SNR (for a 24 bit data word). However, even 123dB works out to an effective bit length of (123-1.76)/6 = 20 bits, not 24 bits. Still, better than the SGTL by quite a lot. The PCM4220 can sample at 192KSPS, as well. Unlike the SGTL5000 the PCM4220's operating modes are all set up using device pins -- so you actually could hardwire the thing for a specific application and not have to bother with an I2C or SPI interface. Of course, your processor still has to support an I2S serial input stream.
On DSOs it is possible to put them into an averaging mode, where the acquired data is averaged over time -- and this greatly improves the effective SNR. However, pulses coming from a scintillator are 1-shot events so it's not possible to use averaging. I thought about this some but couldn't come up with a way to perform averaging without mangling the multi-channel analyzer function. It doesn't appear to be easy to do because you'd have to classify each pulse correctly BEFORE you average it with a previously-identified pulse. Sort of a nasty chicken and egg paradox there. However, if you generate a pulse "library" by using pure elemental samples you MIGHT be able to do some meaningful classifications. Something to think about.....
This is why I'm looking at the PCM4220 which boasts a 123dB SNR (for a 24 bit data word). However, even 123dB works out to an effective bit length of (123-1.76)/6 = 20 bits, not 24 bits. Still, better than the SGTL by quite a lot. The PCM4220 can sample at 192KSPS, as well. Unlike the SGTL5000 the PCM4220's operating modes are all set up using device pins -- so you actually could hardwire the thing for a specific application and not have to bother with an I2C or SPI interface. Of course, your processor still has to support an I2S serial input stream.
On DSOs it is possible to put them into an averaging mode, where the acquired data is averaged over time -- and this greatly improves the effective SNR. However, pulses coming from a scintillator are 1-shot events so it's not possible to use averaging. I thought about this some but couldn't come up with a way to perform averaging without mangling the multi-channel analyzer function. It doesn't appear to be easy to do because you'd have to classify each pulse correctly BEFORE you average it with a previously-identified pulse. Sort of a nasty chicken and egg paradox there. However, if you generate a pulse "library" by using pure elemental samples you MIGHT be able to do some meaningful classifications. Something to think about.....
You somehow track my thoughts.
I have trawled available ADC's trying to select on speed, number of bits resolution, signal-noise ratio, in-stock availability, price, and signal processing complexity overhead. I come a number of conclusions that reduce the choices. I throw in a little preference from previous actual design implementation and experience of trying to reduce the number of lower order bits that were being rendered random by (mostly) common-mode noise from SMPUs, computing, the isolation in transfer of the digital numbers from the point where there is necessarily a common ground return. At the time, this was to measure S-Band Satcoms transmitter over a 76dB dynamic range.
1. The A/D converter choice converges on Analog Devices, and Linear Technology products, and they are now all part of the same company.
2. I cut the choice by looking for only the products in stock at Mouser and Digi-Key. There is some overlap with Farnell/Element-14 and RS Components. The rest of the choice is driven by price. I was OK with under $10 to $30.
The whole thing, built as a 16-bit, 5MB/sec rate device, on demo PCB seems to cost around £44 to £55. (Sorry - I freely use both $ and £, and I don't know where sits "Mackenzie Kings") Maybe too pricey here, after the other costs. Makes me want to go after "pulse stretching"
3. I was pleased to discover that of the possible products, most of the range of Linear Technology part numbers, now mixed in with some Analg Devices part numbers have available evaluation boards. The designs are published, or you can buy some PCB only. Also, can buy the full assembled kit. These can come in two forms. One includes an elaborate interface, with more electronics, and cable to PC, with Windows "evaluation" software. Then there is the simpler kind, which is the device, and some surrounding isolation interface parts, like voltage level translators, local low noise regulators, etc. Intended for I2C and SPI interfaces.
4. You are right about the number of bits, but take care. When the product claims "no lost codes", what may be going on is that to deliver (say) 16 bits, the LSB bit is actually a 17th bit. The normal LSB uncertainty over where the threshold is (turning 65536-1 levels into 32768-1 levels) is removed. That extra bit hidden away in the bottom register makes the whole 16-bits available.
Mixed in there is consider the sampling method. For example, the Delta-Sigma types get past quite a lot of quantization noise. This is where you have to spot what are 1-bit modulators. We don't go into the principles here, but there is a speed trade-off in return for signal quality. The implementation can offer cost convenience and reduced circuit complexity - but slower.
Some A/D converters get up to fast types of pipelined averaging, which boosts signal-to-noise ratio.
4a. I consider the signal-to-noise ratio available from Si(PM) avalanche diodes in reverse bias at room temperatures, and parallel thoughts about PMT shot noise, dynode noise modulation from supplies, photocathode dark current, etc. There is no point in having a 90dB S/N A/D converter when the lower order bits are jumping about from a incoming 60dB S/N ratio. In passing, S/N ratios for me are power ratios..
10Log10(signal_power/noise_power).
5. I throw in considering using the filtered pulse stretching method, which has to get a credibility check. Even if it works well, the circuit implementations published, especially the filter, are not going to be followed by me.
I would "roll my own".
This opens the possibility of offering buffered, low noise, sorted pulses as line-level inputs into the audio A/D channels of the very same Pi doing the rest of the work.
6. "Pulse Stretching"
After what I have read about scintillators, afterpulses, etc, I am definitely a fan of high speed A/D conversion intitial information snatching.
All that stuff about "losing the ringing", caused by the filtering in the first place, does not impress me. What amounts to a crude diode clamping instead of an analog instrumentation circuit - what? All that stuff to deal with an artifact needlessly the consequence of strange signal conditioning in the first place! I am willing to check out the validity of "pulse stretching", but in a deliberate way to prove it works.
If the amplitude information is retained, and the only distortions are from constants in snatching it into energy storage in a filter, and waiting out the aftermath to leak away before bumping it again , then OK. The circuit needs to be gated, and block A/D conversion until ready for the "next" pulse candidate.
Triggering!
The random arrival of pulses does not prevent my DSO scope from averaging. If the trigger is set high enough to to miss the low-level pulses, it builds a trace, triggering when it has to. It remains to be tried out on a proper scintillation signal.
Forgive that all this is a bit expansive. I wanted to let you know my thinking, and where I am as yet undecided.
The stuff from Sphere's is interesting --> https://www.sphere.bc.ca/test/photo-tubes.html
So far, the shipping $66 exceeds some of the product $45.
My "other" PMT is somewhere "in transit" from Ukraine.
---------------
Do you have a Si(PM) photodiode in mind? How much does it cost?
What is your (so far) choice for scintillator?
Are we going to attempt an acryllic " long pyramid" light pipe to deliver from scintillator to a 1.4mm2 diode?
It's quite strange - shovelling shop site spoil into a heap (energetic + boring) while thinking about A/D conversion!
I have trawled available ADC's trying to select on speed, number of bits resolution, signal-noise ratio, in-stock availability, price, and signal processing complexity overhead. I come a number of conclusions that reduce the choices. I throw in a little preference from previous actual design implementation and experience of trying to reduce the number of lower order bits that were being rendered random by (mostly) common-mode noise from SMPUs, computing, the isolation in transfer of the digital numbers from the point where there is necessarily a common ground return. At the time, this was to measure S-Band Satcoms transmitter over a 76dB dynamic range.
1. The A/D converter choice converges on Analog Devices, and Linear Technology products, and they are now all part of the same company.
2. I cut the choice by looking for only the products in stock at Mouser and Digi-Key. There is some overlap with Farnell/Element-14 and RS Components. The rest of the choice is driven by price. I was OK with under $10 to $30.
The whole thing, built as a 16-bit, 5MB/sec rate device, on demo PCB seems to cost around £44 to £55. (Sorry - I freely use both $ and £, and I don't know where sits "Mackenzie Kings") Maybe too pricey here, after the other costs. Makes me want to go after "pulse stretching"
3. I was pleased to discover that of the possible products, most of the range of Linear Technology part numbers, now mixed in with some Analg Devices part numbers have available evaluation boards. The designs are published, or you can buy some PCB only. Also, can buy the full assembled kit. These can come in two forms. One includes an elaborate interface, with more electronics, and cable to PC, with Windows "evaluation" software. Then there is the simpler kind, which is the device, and some surrounding isolation interface parts, like voltage level translators, local low noise regulators, etc. Intended for I2C and SPI interfaces.
4. You are right about the number of bits, but take care. When the product claims "no lost codes", what may be going on is that to deliver (say) 16 bits, the LSB bit is actually a 17th bit. The normal LSB uncertainty over where the threshold is (turning 65536-1 levels into 32768-1 levels) is removed. That extra bit hidden away in the bottom register makes the whole 16-bits available.
Mixed in there is consider the sampling method. For example, the Delta-Sigma types get past quite a lot of quantization noise. This is where you have to spot what are 1-bit modulators. We don't go into the principles here, but there is a speed trade-off in return for signal quality. The implementation can offer cost convenience and reduced circuit complexity - but slower.
Some A/D converters get up to fast types of pipelined averaging, which boosts signal-to-noise ratio.
4a. I consider the signal-to-noise ratio available from Si(PM) avalanche diodes in reverse bias at room temperatures, and parallel thoughts about PMT shot noise, dynode noise modulation from supplies, photocathode dark current, etc. There is no point in having a 90dB S/N A/D converter when the lower order bits are jumping about from a incoming 60dB S/N ratio. In passing, S/N ratios for me are power ratios..
10Log10(signal_power/noise_power).
5. I throw in considering using the filtered pulse stretching method, which has to get a credibility check. Even if it works well, the circuit implementations published, especially the filter, are not going to be followed by me.
I would "roll my own".
This opens the possibility of offering buffered, low noise, sorted pulses as line-level inputs into the audio A/D channels of the very same Pi doing the rest of the work.
6. "Pulse Stretching"
After what I have read about scintillators, afterpulses, etc, I am definitely a fan of high speed A/D conversion intitial information snatching.
All that stuff about "losing the ringing", caused by the filtering in the first place, does not impress me. What amounts to a crude diode clamping instead of an analog instrumentation circuit - what? All that stuff to deal with an artifact needlessly the consequence of strange signal conditioning in the first place! I am willing to check out the validity of "pulse stretching", but in a deliberate way to prove it works.
If the amplitude information is retained, and the only distortions are from constants in snatching it into energy storage in a filter, and waiting out the aftermath to leak away before bumping it again , then OK. The circuit needs to be gated, and block A/D conversion until ready for the "next" pulse candidate.
Triggering!
The random arrival of pulses does not prevent my DSO scope from averaging. If the trigger is set high enough to to miss the low-level pulses, it builds a trace, triggering when it has to. It remains to be tried out on a proper scintillation signal.
Forgive that all this is a bit expansive. I wanted to let you know my thinking, and where I am as yet undecided.
The stuff from Sphere's is interesting --> https://www.sphere.bc.ca/test/photo-tubes.html
So far, the shipping $66 exceeds some of the product $45.
My "other" PMT is somewhere "in transit" from Ukraine.
---------------
Do you have a Si(PM) photodiode in mind? How much does it cost?
What is your (so far) choice for scintillator?
Are we going to attempt an acryllic " long pyramid" light pipe to deliver from scintillator to a 1.4mm2 diode?
It's quite strange - shovelling shop site spoil into a heap (energetic + boring) while thinking about A/D conversion!
I also am looking at DigiKey as the main parts supplier. They have SiPM eval boards for not much more than the bare chip so for a first pass I'll go that direction. So far I haven't found anyone selling SiPM boards on ebay.
As far as a scintillator goes, I have a couple of LYSO crystals that can be used to at least evaluate a complete DAQ system and also permit experimenting with better-designed pulse stretching circuits. Right now I'm not going to try a light pipe, will initially try wrapping the sides of the crystal with aluminum foil to direct more photons into the SiPM. Some papers I've read have described wrapping white teflon tape around the crystals or painting them with highly reflective white paint.
The issue with using a DSO for averaging is that its triggering isn't all that sophisticated when it comes to distinguishing an iron pulse from a nickel pulse -- the energy difference is just barely noticeable, if you believe the Theremino results. So if you are averaging iron and nickel pulses together you aren't going to get very far.
As far as a scintillator goes, I have a couple of LYSO crystals that can be used to at least evaluate a complete DAQ system and also permit experimenting with better-designed pulse stretching circuits. Right now I'm not going to try a light pipe, will initially try wrapping the sides of the crystal with aluminum foil to direct more photons into the SiPM. Some papers I've read have described wrapping white teflon tape around the crystals or painting them with highly reflective white paint.
The issue with using a DSO for averaging is that its triggering isn't all that sophisticated when it comes to distinguishing an iron pulse from a nickel pulse -- the energy difference is just barely noticeable, if you believe the Theremino results. So if you are averaging iron and nickel pulses together you aren't going to get very far.
OK - if you think there may be advantage that we both go down the same road starting with the same Si(PM) then I can get the same one as you have. I will of course be attempting things with the PMT. PMTs appear able to outperform Si(PM) in various ways, but require very careful, specific setups to do that, and have a big "looking after it" overhead. They die from runaway electron burn inadvertent light leak, and even need to be stored in the dark.
Should I start with the same Si(PM), or should we explore a couple of different ones.?
Which LYSO did you get? I will have to trawl the thread previous posts to find the link to that catalogue again.
OK - agreed. At least for design, any DSO captured pulse is useful in the early stages.
For distinguishing iron from nickel, that is a challenge, because Ni is such a common alloy add-in to iron.
Come to that, most steel now that has come with a considerable recycled component tends to have at least some of a whole lot of elements as remnants, including a little nickel. Steel is even getting slightly more radioactive over time from recycled input.
Re: Pulse Stretcher
I still don't quite get it why all that was ever necessary. It is only that the guys who did that design were plenty smart, actually made the kit, and put a lot of research into it, and wrote a lot about it, has me be cautious. I keep thinking I might have missed something.
If the signal from nickel is hard to distinguish from that of iron, then it is hardly likely to help that the signal be smeared by having a first differentiator without op-amp assist nor buffering (i.e. the low value series capacitor), and then the remnant of energy be low-pass filtered such that the transient attempts to charge whatever filter components store energy to finally leak away slowly, making a "stretched" pulse that has a (tiny) maximum amplitude vaguely related to the the height of the original pulse. Any difference of Ni to Fe signals is now a whole lot smaller!
If all we need is the measured height of the peak, let us snatch it with a peak detector. These can work even at microwave speeds, though we need hardly go that far. Then, we can can measure the level "at leisure" with an affordable, accurate A/D, before "reset", and ignoring the remainder of the pulse.
Cool thread!
I do believe I got left at the curb around Post #2...
and here I thought this might be a nice long thread on getting lawn equipment running.
I do believe I got left at the curb around Post #2...
and here I thought this might be a nice long thread on getting lawn equipment running.
Sure Dan - good that you still look in.Cool thread!
I do believe I got left at the curb around Post #2...
and here I thought this might be a nice long thread on getting lawn equipment running.
This is going to be a fairly determined experiment, but I have not lost sight of the hope that it can maybe yield an affordable gadget/app thing that can help tell what the steel is.
Sometimes I am thinking that I may have a better shot at lawn equipment!
I'm not so sure about that. I believe that your endeavour in creating a test for materials is a much better effort that trying to keep lawn eq running.Sure Dan - good that you still look in.
This is going to be a fairly determined experiment, but I have not lost sight of the hope that it can maybe yield an affordable gadget/app thing that can help tell what the steel is.
Sometimes I am thinking that I may have a better shot at lawn equipment!
Currently I have (4) gas trimmers, and a chainsaw not running.
(Sorry about the off-topic, but I couldn't resist...
)
I've been debating over the 3x3mm or 6x6mm SiPM -- see this list on Digikey (please let me know if the link doesn't work). The 3x3 is a little too small for the LYSO scintillators I got here, which are 4x4mm on the SiPM face. If I want to avoid the complication of a light pipe I guess that means the more expensive 6x6, at about $100USD. The bare chips are darn near the same price, something to think about if you're going to lay out a PCB for an ADC anyway.
Looking at the Theremino approach, it appears to me that the main thrust of their approach is twofold. The first is noise management using the passive filters. I'm thinking that a Bessel low pass filter instead of a crappy passive "whatever" filter might be beneficial, because it will do a better job of preserving the pulse shape. This said, an active filter that can handle sub-microsecond input pulses suggests a pretty fast amplifier will be needed in order to really achieve the filter performance you want. The equivalent of a 741 op-amp will NOT cut it. This is where LTC's Spice package can really come in handy, since it includes models for a number of their amplifiers. I have Wine, guess I need to download LTC Spice.
The second Theremino approach is pulse management, where they cull out all the doubtful pulses . They also claim some improvements using their custom deconvolution S/W but that's where I would want to use either Octave or SciPy and leverage stuff that's already out there.
I've gotta say this is is a really interesting project, if that wasn't obvious
. It uses a number of different disciplines that I really like, and the end result depends on balancing physics, optics and electronics (and an underlying $ consideration) that's so much fun. Really a prototypical definition of what "hobbyist" means in the best sense of the word. The broad range of knowledge of folks that have contributed to this thread is terrific, and, best yet, I've learned some things along the way. What's not to like about that!
Looking at the Theremino approach, it appears to me that the main thrust of their approach is twofold. The first is noise management using the passive filters. I'm thinking that a Bessel low pass filter instead of a crappy passive "whatever" filter might be beneficial, because it will do a better job of preserving the pulse shape. This said, an active filter that can handle sub-microsecond input pulses suggests a pretty fast amplifier will be needed in order to really achieve the filter performance you want. The equivalent of a 741 op-amp will NOT cut it. This is where LTC's Spice package can really come in handy, since it includes models for a number of their amplifiers. I have Wine, guess I need to download LTC Spice.
The second Theremino approach is pulse management, where they cull out all the doubtful pulses . They also claim some improvements using their custom deconvolution S/W but that's where I would want to use either Octave or SciPy and leverage stuff that's already out there.
I've gotta say this is is a really interesting project, if that wasn't obvious
. It uses a number of different disciplines that I really like, and the end result depends on balancing physics, optics and electronics (and an underlying $ consideration) that's so much fun. Really a prototypical definition of what "hobbyist" means in the best sense of the word. The broad range of knowledge of folks that have contributed to this thread is terrific, and, best yet, I've learned some things along the way. What's not to like about that!