Concept Design

[Willy Bayot]

[QUOTE=Olly;n410208]

Make sure not to confuse sampling frequency Fsmpl with the frequency of the input signal being measured Finput.

Regards

I just wanted to signal that the text at the beginning of the data sheet did not define that the SNR was measured at fIN of 2Khz only.
We need more on our application and there is no graph showing how it varies with fIN larger than 2Khz while there is such a graph on the datasheet of the MAXIM component.

[Willy Bayot]

create a shortlist of the candidate ADC ... here is my vote LTC2338-18

Looks nice but stock is low (37 at Mouser) and steep price ($56)

[Olly]

[QUOTE=Willy Bayot;n410214]

I just wanted to signal that the text at the beginning of the data sheet did not define that the SNR was measured at fIN of 2Khz only.
We need more on our application and there is no graph showing how it varies with fIN larger than 2Khz while there is such a graph on the datasheet of the MAXIM component.

Yes, I'm afraid there is a lot of gamesmanship played by the manufacturers when producing datasheets and examining the graphs is almost always necessary in order to get a true picture of the performance.

Normally SNR is not really a function of input frequency as the noise floor is generally measured in the absence of a signal or with the input shorted.
SINAD and ENOB however are very much a function of input frequency as these are dependant on the distortion (THD) of the device.

Looking at the graphs it would appear that the MAX11158 has the worst THD specs at 100kHz input signal (-90dB) whereas the LTC2376-18 achieves -115dB THD with 100kHz input which on the face of it is exceptional but the reality is that it only samples at 250kHz so any harmonics above the Nyquist frequency of 125kHz are pretty much shifted out of the measurement band thus improving the measured THD. Even so, it's THD specs are considerably better than the MAX11158 and if a sample rate of 250kHz is fast enough then this is the device I would choose.

If a faster sample rate is required then I would stick with the LTC2380-24 as its THD at 100kHz is still 15dB better than the Maxim part which should translate to better SINAD and ENOB.

Just my 2c worth
Regards

[Willy Bayot]

[QUOTE=Olly;n410216]

Yes, I'm afraid there is a lot of gamesmanship played by the manufacturers when producing datasheets and examining the graphs is almost always necessary in order to get a true picture of the performance.

Normally SNR is not really a function of input frequency as the noise floor is generally measured in the absence of a signal or with the input shorted.
SINAD and ENOB however are very much a function of input frequency as these are dependant on the distortion (THD) of the device.

Looking at the graphs it would appear that the MAX11158 has the worst THD specs at 100kHz input signal (-90dB) whereas the LTC2376-18 achieves -115dB THD with 100kHz input which on the face of it is exceptional but the reality is that it only samples at 250kHz so any harmonics above the Nyquist frequency of 125kHz are pretty much shifted out of the measurement band thus improving the measured THD. Even so, it's THD specs are considerably better than the MAX11158 and if a sample rate of 250kHz is fast enough then this is the device I would choose.

If a faster sample rate is required then I would stick with the LTC2380-24 as its THD at 100kHz is still 15dB better than the Maxim part which should translate to better SINAD and ENOB.

Just my 2c worth
Regards

Thanks, that's the recommendation we expected from an expert.
Thus, let's continue on the LTC23xx track if nobody has a better proposal.

[Carl-NC]

I saw several times mentioning high gain and I don't understand why?
Stong TX and high gain at RX = always not winning combination, imho.

My original noise calculations assumed a gain of 25 as per Tony's LT-Spice circuit. That's too low. OTOH, most PI designs run the preamp gain at 500-1000 and that's probably too high. I'm initially considering 100-200 for a starting point. You want to run the gain as high as possible without overloading the preamp. Higher gain increases the output noise but also increases the output signal at the same rate. Depending on where the gain is applied (preferably in the first stage), higher gain can actually improve SNR.

Addendum: the highest gain we can actually run will depend on the worst ground signal encountered and how that interacts with the square wave TX. I don't have a firm grasp of what that will look like so for now choosing the gain is mostly a guess.
------------------------------

On the ADC, it is important to understand what performance metrics matter and which do not. First, we are sampling a wideband signal that exceeds the Nyquist BW of the ADC, even if we use 1MSps. Way back I casually stated that a 0.25us nugget (sub-grain for sure) needed at least 636kHz of BW. Technically an exponential decay has infinite BW and 1/(2π∙τ) captures half of the energy. We would actually prefer more BW but I'm concerned that excess noise will swamp out weak signals. So if we set up a Nyquist filter like you would normally do (fs/2), we lose small nuggets. So our ADC will be folding in some of the second Nyquist zone noise which is why I'm especially concerned about SNR.

Also, SNR is a function of frequency because of the clock jitter in the SHA. In fact, except for quantization noise the SHA is the most noise-dominant element in the ADC. The raw size of the hold cap determines the thermal noise (the so-called kT/C noise) and the clock jitter determines the frequency degradation. You would normally design the clock for minimal degradation over the first Nyquist band unless you are intentionally designing an IF-sampling ADC. The last ADC I designed for Analog Devices was the AD6645, here is the SNR-vs-f plot:


This ADC was designed for IF sampling so even though the clock is 80MHz the SNR is plotted out to 200MHz. The roll-off is almost entirely due to jitter.

Harmonics don't matter so much for the AMX. Harmonics give rise to signal distortion but we are not so much concerned with pure fidelity because there are probably no pure signal responses in the first place. Nuggets tend to have multiple domain responses where a single exponential dominates, so it's already a distorted signal. Viscous ground is a power-law response with some variation in the power term so it's also a non-pure response. Even a -60dB spur in the ADC is only 0.1%, I think we can live with that. Also, harmonics internally generated by the ADC (which are what you are seeing in the data sheet) are not limited to the first Nyquist zone. They are infinite in BW and fold back into the first Nyquist zone. Again, the AD6645 shows harmonics out to 200MHz even when sampling at 80MHz:



In cases where SNR is not clearly plotted over f the next best thing is SINAD-vs-f (which simply includes the harmonic energy) or the ENOB-vs-f (which should be directly derived from SINAD). Again, THD and SFDR are of little interest in this application.

Now, all that said, which ADC are we using? I think I should revisit my noise calculations with a more reasonable gain.

[ivconic]

Thanks Carl, that clarified some misty areas in my understandings.
I quickly reviewed some tables and I can draw some general conclusions.
It is logical that the effective number of bits will decrease with the increase of data rate and gain.
It is quite a complicated and intricate relationship.
In order to successfully resolve this; first we need to determine how many bits we really need, expressed as the number of effective bits.
For example if our work ends well with 16 bits; obviously we will need an ADC that has 16 + n bits discarded due to SNR and distortion.
But it would be good and simpler if the whole process happens in the ADC itself and not afterwards.
Therefore, I would change the criteria a little and as the first I would put an ADC with a programmable gain and internal programmable low-pass digital filters.
It is clear that as more as possible bits is preferable, because of the above some number of bits will have to be discarded and we will be left with effective bits available.
High demands on initial Tony's wish list obviously leads to even higher demands for such ADC.
I think we got involved in some pretty dubious discussions and thoughts with this.
And I still don't have a clear concept.
I suggest to start with something and to change the parts as you go, if necessary.
Because finding the ideal ADC in this way, with such dubious analyses... will lead to the same conclusion in the end; that there is no ideal ADC and that by choosing any ADC
will also have to accept certain tradeoffs.
The best would be an ADC that has everything in it. Programmable gain, a series of filters and its own internal reference. I'd put sample rate before bitrate too.

[ivconic]

Otherwise, it should not be surprising that manufacturers always put "big numbers" in their specifications. Because it attracts attention and customers.
It's a trading part of the story.
They put 24 bits in the title and later in the analysis, they often explain in "small letters" that the number of effective bits is much smaller.
Justification is easy for them; it is a general purpose ADC and in some cases it will fill in what is written. (by the Murphy's law it is never our case on that list)
That's why I have a much shorter evaluation method. Rule of thumb. Instead of fumbling with tons of calculations.
If it says 24 bits, the rule of thumb tells me it's actually 18-20 effective bits.Often less than that.
That's why sample rate is more important to me than bitrate.In cases like this PI project.
16 effective bits is about enough to fulfill the wishes on Tony's list.
But it is much more important and useful that the ADC has everything on it.
...
Calculation of gain and how to determine the right gain?
First of all, the ADC should have an adjustable gain. (fancy word is PGA)
And it's never productive to use extreme values.
If the maximum is 128; always choose half of that, so 64x.
Based on that value, it should now be easier for you to calculate the required gain on the two-stage RX frontend?

[ivconic]

I believe in the success of this project. With a little good will, everything can be put together in a perfect combination.

[Carl-NC]

Therefore, I would change the criteria a little and as the first I would put an ADC with a programmable gain and internal programmable low-pass digital filters.

Hardly any of the single channel ADCs have a PGA or filters, they are more common on multi-channel ADCs. Even then, I am usually not impressed with the performance of the internal PGA. I can do better by designing the correct gain outside the ADC. Built-in filters also have limited use because we are using the ADC with a wideband signal that exceeds the first Nyquist zone. So I don't consider either of these features useful in this application.

Because finding the ideal ADC in this way, with such dubious analyses... will lead to the same conclusion in the end; that there is no ideal ADC and that by choosing any ADC
will also have to accept certain tradeoffs.

This is called "engineering." In a cutting-edge design it is critically important to be able to do these analyses so you know where the limitations are, otherwise you're just guessing. That's why I did a noise analysis on the AFE, so I would know what ADC SNR is required.

[ivconic]

Hardly any of the single channel ADCs have a PGA or filters, they are more common on multi-channel ADCs. Even then, I am usually not impressed with the performance of the internal PGA. I can do better by designing the correct gain outside the ADC. Built-in filters also have limited use because we are using the ADC with a wideband signal that exceeds the first Nyquist zone. So I don't consider either of these features useful in this application.
This is called "engineering." In a cutting-edge design it is critically important to be able to do these analyses so you know where the limitations are, otherwise you're just guessing. That's why I did a noise analysis on the AFE, so I would know what ADC SNR is required.

In both of your comments is obvious that your practical exepriences are of great value here, simply; you tried it and you know. I can only assume relying on what I can read in the documentation... and be wrong, of course.
Ok, now more things are clear.
...
Next thing to ask;
this:




and the specs:

Board Specifications

• RP2040 microcontroller chip designed by Raspberry Pi in the United Kingdom
• Dual-core Arm Cortex M0+ processor, flexible clock running up to 133 MHz
• 264KB of SRAM, and 2MB of on-board Flash memory
• USB-C connector, keeps it up to date, easier to use
• Castellated module allows soldering direct to carrier boards
• USB 1.1 with device and host support
• Low-power sleep and dormant modes
• Drag-and-drop programming using mass storage over USB
• 29 × multi-function GPIO pins (20× via edge pinout, others via solder points)
• 2 × SPI, 2 × I2C, 2 × UART, 4 × 12-bit ADC, 16 × controllable PWM channels
• Accurate clock and timer on-chip
• Temperature sensor
• Accelerated floating-point libraries on-chip
• 8 × Programmable I/O (PIO) state machines for custom peripheral support

C/C++，MicroPython Support
Comprehensive SDK, Dev Resources, Tutorials To Help You Easily Get Started
Dual-Core Arm Processor
Dual-Core Arm Cortex M0+ Processor, Flexible Clock Running Up To 133 MHz
29 × Multi-Function GPIO Pins
Configurable Pin Function, Allows Flexible Development And Integration

How about it?
Can it be the "hearth" of the detector?
I don't have any exeprience with such one, just stumbled on it in local adverts.
And the price is 7e !!!!

[Olly]

In both of your comments is obvious that your practical exepriences are of great value here, simply; you tried it and you know. I can only assume relying on what I can read in the documentation... and be wrong, of course.
Ok, now more things are clear.
...
Next thing to ask;
this:




and the specs:

Board Specifications

• RP2040 microcontroller chip designed by Raspberry Pi in the United Kingdom
• Dual-core Arm Cortex M0+ processor, flexible clock running up to 133 MHz
• 264KB of SRAM, and 2MB of on-board Flash memory
• USB-C connector, keeps it up to date, easier to use
• Castellated module allows soldering direct to carrier boards
• USB 1.1 with device and host support
• Low-power sleep and dormant modes
• Drag-and-drop programming using mass storage over USB
• 29 × multi-function GPIO pins (20× via edge pinout, others via solder points)
• 2 × SPI, 2 × I2C, 2 × UART, 4 × 12-bit ADC, 16 × controllable PWM channels
• Accurate clock and timer on-chip
• Temperature sensor
• Accelerated floating-point libraries on-chip
• 8 × Programmable I/O (PIO) state machines for custom peripheral support

C/C++，MicroPython Support
Comprehensive SDK, Dev Resources, Tutorials To Help You Easily Get Started
Dual-Core Arm Processor
Dual-Core Arm Cortex M0+ Processor, Flexible Clock Running Up To 133 MHz
29 × Multi-Function GPIO Pins
Configurable Pin Function, Allows Flexible Development And Integration

How about it?
Can it be the "hearth" of the detector?
I don't have any exeprience with such one, just stumbled on it in local adverts.
And the price is 7e !!!!

The full size Raspberry Pi Pico is even cheaper at around £4 each. These are great little boards and I’ve used them quite successfully as controllers for PI detectors. At that price you could use more than 1 in a project e.g one for display and 1 for processing etc. The programmable PIOs are quite unique and allow very fast and flexible custom interfaces which run independently of the 2 main cores.

All in all, I like these a lot.

[ivconic]

Missed to see the details... how many timers on it?
Yes there is also full sized module on local adverts, price is almost the same here, affordable for sure.

[ivconic]

As for the RF frontend... the opamps! Headache!
Yet I located one pretty interesting at local supplier and by the price of 3.7e.
It is OPA350PA...
What's even better (for me) it is in DIL package. So I can make a through hole variant for my experiments.
For me it sounds like ideal opamp for this use!
Opinions?

[Olly]

Missed to see the details... how many timers on it?
Yes there is also full sized module on local adverts, price is almost the same here, affordable for sure.

It doesn’t have conventional timers as such but has up to 16 PWM channels and 8 programmable PIO state machines which will probably be ideal for creating the dual or triple frequency TX pulses

[ivconic]

As for the RF frontend... the opamps! Headache!
Yet I located one pretty interesting at local supplier and by the price of 3.7e.
It is OPA350PA...
What's even better (for me) it is in DIL package. So I can make a through hole variant for my experiments.
For me it sounds like ideal opamp for this use!
Opinions?

- RAIL-TO-RAIL INPUT
- RAIL-TO-RAIL OUTPUT (within 10mV)
- WIDE BANDWIDTH: 38MHz
- HIGH SLEW RATE: 22V/µs
- LOW NOISE: 5nV/√Hz
- LOW THD+NOISE: 0.0006%
- UNITY-GAIN STABLE
- MicroSIZE PACKAGES
- SINGLE, DUAL, AND QUAD
- optimized for low voltage, single-supply operation

High speed operation (38MHz, 22V/µs) make them ideal for driving sampling Analog-to-Digital (A/D) converters.
The OPA350 series operates on a single supply as low as 2.5V (up to 7V).