I'm considering trying to build an IoT soundlevel meter and wondering if modern 24bit ADCs can support such a use-case.
I'm hoping for a dynamic range of about 120dB, and power draw around 10-20mW (just for the ADC not including draw from the driver and other components). Not too sure about other specs yet.
I saw the upcoming ad4630 which with about 20 averages and scaled down CNV clock gets within the ballpark I think. Prototyping with that package doesn't look fun though and I don't know when the part will be available.
LTC2500 looks like it could be an option, again with some use of averaging.
Anything else I should consider?
If this is my first ever ADC design project and I'm also pretty green with MCUs should I start with an easier task and work up to this? I was eyeing the older AD7766; although it won't meet my specs it looks a bit more simple and like I may be able to use it with an Arduino fairly easily. Suggestions welcome.
I'm hoping for a dynamic range of about 120dB, and power draw around 10-20mW (just for the ADC not including draw from the driver and other components). Not too sure about other specs yet.
I saw the upcoming ad4630 which with about 20 averages and scaled down CNV clock gets within the ballpark I think. Prototyping with that package doesn't look fun though and I don't know when the part will be available.
LTC2500 looks like it could be an option, again with some use of averaging.
Anything else I should consider?
If this is my first ever ADC design project and I'm also pretty green with MCUs should I start with an easier task and work up to this? I was eyeing the older AD7766; although it won't meet my specs it looks a bit more simple and like I may be able to use it with an Arduino fairly easily. Suggestions welcome.
The “ideal” snr improvement = sqrt(number of measurements) from what I’ve read.. take with a pinch of salt.
There are complications - such as oversampling the same sample or averaging the samples to simulate the same. This also assumes a random noise spread and the noise isn’t correlated to the signal.
In astrophotography you take multiple long exposures and then align the stars so that the image sits on top. That correlation of the signal (aligning using the strong signals of the stars) then causes the feature signal to appear stronger. If a pixel is stuck at full brightness then the image stacking will show a bright pixel too - an example of noise correlated to the measurement that is another issue that won’t be removed).
4dB = sqrt(16) however you will find the measurement noise larger plus the sound through a speaker may not correlate that well. (I’ve used FFT phase correlation to align images and focusing - you could phase align a sine wave but the complexity of a room sound from the signal may prove difficult if hand held). So 45 samples averaged would be 6.7dB so that sounds like a reasonable start.
It’s also worth looking at the specs - you will need at least 2x the bandwidth in sample rate. With any filters setup, you may find that you have less “oversampling” available.
There are complications - such as oversampling the same sample or averaging the samples to simulate the same. This also assumes a random noise spread and the noise isn’t correlated to the signal.
In astrophotography you take multiple long exposures and then align the stars so that the image sits on top. That correlation of the signal (aligning using the strong signals of the stars) then causes the feature signal to appear stronger. If a pixel is stuck at full brightness then the image stacking will show a bright pixel too - an example of noise correlated to the measurement that is another issue that won’t be removed).
4dB = sqrt(16) however you will find the measurement noise larger plus the sound through a speaker may not correlate that well. (I’ve used FFT phase correlation to align images and focusing - you could phase align a sine wave but the complexity of a room sound from the signal may prove difficult if hand held). So 45 samples averaged would be 6.7dB so that sounds like a reasonable start.
It’s also worth looking at the specs - you will need at least 2x the bandwidth in sample rate. With any filters setup, you may find that you have less “oversampling” available.
Last edited:
https://www.analog.com/media/en/technical-documentation/data-sheets/250032fb.pdf
Page 30 shows sample rate, bandwidth, over sampling and resulting Dynamic range and noise.
If a filter is used you’re already starting to loose sample rate and bandwidth.
You could use multiple ADCs with the same clock then combine the samples as long as they both correlate.
Page 30 shows sample rate, bandwidth, over sampling and resulting Dynamic range and noise.
If a filter is used you’re already starting to loose sample rate and bandwidth.
You could use multiple ADCs with the same clock then combine the samples as long as they both correlate.
Last edited:
OK so I'm anxious to get my hands dirty and start learning. Having read about a million datasheets I'm leaning towards exploring the LTC2500-32 despite the fact it seems ambitious.
I'm not sure I'm ready to invest in the EVAL board and I'm wondering if I can start to work with this chip somewhat inexpensively without it. As a first step I was thinking of soldering to a DIP-24 breakout board and experimenting on a solderless breadboard, having no illusions about getting performance near what is spec'd.
I was thinking a first goal would be just to power & initialize the chip and read sample values to an Arduino with the analog inputs held to VREF/2. Am I likely to even get this far?
I figure parasitic capacitance from a solderless breadboard shouldn't be an issue at a 2MHz sample rate, however I noticed in the datasheet it states MCLK should be held high for only 40ns. That's 8% duty cycle of the sample rate which I guess means I need a 25MHz clock source - isn't that going to be a problem?
I don't have a bench clock source, I was hoping I could use a clock generator IC but I'm wondering if this puts things out of solderless breadboard territory.
I'm not sure I'm ready to invest in the EVAL board and I'm wondering if I can start to work with this chip somewhat inexpensively without it. As a first step I was thinking of soldering to a DIP-24 breakout board and experimenting on a solderless breadboard, having no illusions about getting performance near what is spec'd.
I was thinking a first goal would be just to power & initialize the chip and read sample values to an Arduino with the analog inputs held to VREF/2. Am I likely to even get this far?
I figure parasitic capacitance from a solderless breadboard shouldn't be an issue at a 2MHz sample rate, however I noticed in the datasheet it states MCLK should be held high for only 40ns. That's 8% duty cycle of the sample rate which I guess means I need a 25MHz clock source - isn't that going to be a problem?
I don't have a bench clock source, I was hoping I could use a clock generator IC but I'm wondering if this puts things out of solderless breadboard territory.
Last edited:
Worth looking at an ADC driver too perhaps.
If you have a look at the CN0540 board that may give you some ideas. In the end “does it clock?” Well people have build bug style and have working systems. High speed means short runs and being carful of PCB capacitances with ground planes and breadboards etc.
Given mouser has the LTC2500 at £50 each. I would probably try a cheap ADC first with the same form factor..
At a push (just for a prod) you could use a micro controller PWM. Some ADCs will have an inbuilt RC for debugging but clock IC all the way.
If you have a look at the CN0540 board that may give you some ideas. In the end “does it clock?” Well people have build bug style and have working systems. High speed means short runs and being carful of PCB capacitances with ground planes and breadboards etc.
Given mouser has the LTC2500 at £50 each. I would probably try a cheap ADC first with the same form factor..
At a push (just for a prod) you could use a micro controller PWM. Some ADCs will have an inbuilt RC for debugging but clock IC all the way.
Last edited:
Before going to adc extremes look at the input. What microphone will have a useful 120 dB range? And are you ready to commit that level of $$$ into the transducer? It would most likely be a 1/2 inch 200v bias condenser and even then the preamp will be too noisy. I can go into more detail to help narrow this to something that can be made.
