FFT Spectrum Analysis Reviewed

[KE5FX]

I've got a similar oscillator design somewhere here that runs at 6.7GHz and it had similar phase noise performance to the one at 4.4GHz. I've built quite a few oscillators like this and these were used (for example) to test the noise performance of the first few ranges on my HP8566B analyser. The main aim was to make something quick, simple and stable with similar phase noise to the HP8566B.

I don't know if you subscribe to QEX or not, but this might be a good thing to write up for them.

[radar_macgyver]

Those wiggles < 300 Hz could be since you're running off the internal reference, which in an SA44 is a TCXO. Locking to an external stable reference would help get rid of them. This post by hendorog links to an article by Signal Hound where they try to characterize the phase noise of an SA44. You can clearly see the effects of an improved reference oscillator.

Thanks for the writeup. I'll send a link to this thread next time someone brings this up at work I have a hard time explaining why we still need 16-bit A/Ds for receiver front-ends when folks will pull out a data sheet for the latest whizbang 10 GSPS 8-bit A/D and say "we can just filter it with an FPGA".

[hendorog]

One of the reasons for the high noise floor might be the flat top window, which has less side lobe suppression compared to Blackman Harris. Apart from that, I strongly suspect that the ADC might only be 14 bits.

According to SH:
"It’s actually a sigma delta 24 bit ADC, digitally converted to 16-bit I&Q."

https://signalhound.com/support/forums/topic/sa44b-adc-bits-resolution/

[uncle_bob]

One of the reasons for the high noise floor might be the flat top window, which has less side lobe suppression compared to Blackman Harris. Apart from that, I strongly suspect that the ADC might only be 14 bits.

According to SH:
"It’s actually a sigma delta 24 bit ADC, digitally converted to 16-bit I&Q."

https://signalhound.com/support/forums/topic/sa44b-adc-bits-resolution/

Hi

..... which seems like a really silly thing to do if you actually *have* a 24 bit ADC ...

Bob

[hendorog]

One of the reasons for the high noise floor might be the flat top window, which has less side lobe suppression compared to Blackman Harris. Apart from that, I strongly suspect that the ADC might only be 14 bits.

According to SH:
"It’s actually a sigma delta 24 bit ADC, digitally converted to 16-bit I&Q."

https://signalhound.com/support/forums/topic/sa44b-adc-bits-resolution/

Hi

..... which seems like a really silly thing to do if you actually *have* a 24 bit ADC ...

Bob

If that guy said they do, then they do. He is one of the engineers not some marketing rooster.

The ADC was questioned to be 14bits, turns out its 24bits operating effectively at 16bits. Thats a good thing, not a bad thing

Its really only silly if they then spend effort digitally converting back to something greater than 16 bits again isn't it?
 

[uncle_bob]

One of the reasons for the high noise floor might be the flat top window, which has less side lobe suppression compared to Blackman Harris. Apart from that, I strongly suspect that the ADC might only be 14 bits.

According to SH:
"It’s actually a sigma delta 24 bit ADC, digitally converted to 16-bit I&Q."

https://signalhound.com/support/forums/topic/sa44b-adc-bits-resolution/

Hi

..... which seems like a really silly thing to do if you actually *have* a 24 bit ADC ...

Bob

If that guy said they do, then they do. He is one of the engineers not some marketing rooster.

The ADC was questioned to be 14bits, turns out its 24bits operating effectively at 16bits. Thats a good thing, not a bad thing

Its really only silly if they then spend effort digitally converting back to something greater than 16 bits again isn't it?

Hi

I can go out and buy a part that is indeed a 24 bit ADC. The data sheet says 24 bit ADC and at *some* clock rate it operates as a 24 bit ADC. If somebody asks me a question about "what sort of ADC?", I can honestly answer 24 bit ADC.

The gotcha is: It's a part that puts out 24 bits at a 15 Hz sample rate. At a higher rate, it puts out fewer bits. If I'm using it at 100 KHz, it might only be a 12 bit ADC. Since it's only 12 bits, running the rest of the system at 16 is fine.

Bob

[hendorog]

One of the reasons for the high noise floor might be the flat top window, which has less side lobe suppression compared to Blackman Harris. Apart from that, I strongly suspect that the ADC might only be 14 bits.

According to SH:
"It’s actually a sigma delta 24 bit ADC, digitally converted to 16-bit I&Q."

https://signalhound.com/support/forums/topic/sa44b-adc-bits-resolution/

Hi

..... which seems like a really silly thing to do if you actually *have* a 24 bit ADC ...

Bob

If that guy said they do, then they do. He is one of the engineers not some marketing rooster.

The ADC was questioned to be 14bits, turns out its 24bits operating effectively at 16bits. Thats a good thing, not a bad thing

Its really only silly if they then spend effort digitally converting back to something greater than 16 bits again isn't it?

Hi

I can go out and buy a part that is indeed a 24 bit ADC. The data sheet says 24 bit ADC and at *some* clock rate it operates as a 24 bit ADC. If somebody asks me a question about "what sort of ADC?", I can honestly answer 24 bit ADC.

The gotcha is: It's a part that puts out 24 bits at a 15 Hz sample rate. At a higher rate, it puts out fewer bits. If I'm using it at 100 KHz, it might only be a 12 bit ADC. Since it's only 12 bits, running the rest of the system at 16 is fine.

Bob

Yep I understand that - the way you worded your first comment it sounded like you were saying they were lying about the chip being there in the first place.

There is an implication that its not that simple though - I mean just running the ADC at 16bits doesn't really qualify as 'digitally converted' does it?

It would be interesting to know more about how they do it. (Edit: I asked the question in the SH forum so will see what they say)

[uncle_bob]

One of the reasons for the high noise floor might be the flat top window, which has less side lobe suppression compared to Blackman Harris. Apart from that, I strongly suspect that the ADC might only be 14 bits.

According to SH:
"It’s actually a sigma delta 24 bit ADC, digitally converted to 16-bit I&Q."

https://signalhound.com/support/forums/topic/sa44b-adc-bits-resolution/

Hi

..... which seems like a really silly thing to do if you actually *have* a 24 bit ADC ...

Bob

If that guy said they do, then they do. He is one of the engineers not some marketing rooster.

The ADC was questioned to be 14bits, turns out its 24bits operating effectively at 16bits. Thats a good thing, not a bad thing

Its really only silly if they then spend effort digitally converting back to something greater than 16 bits again isn't it?

Hi

I can go out and buy a part that is indeed a 24 bit ADC. The data sheet says 24 bit ADC and at *some* clock rate it operates as a 24 bit ADC. If somebody asks me a question about "what sort of ADC?", I can honestly answer 24 bit ADC.

The gotcha is: It's a part that puts out 24 bits at a 15 Hz sample rate. At a higher rate, it puts out fewer bits. If I'm using it at 100 KHz, it might only be a 12 bit ADC. Since it's only 12 bits, running the rest of the system at 16 is fine.

Bob

Yep I understand that - the way you worded your first comment it sounded like you were saying they were lying about the chip being there in the first place.

There is an implication that its not that simple though - I mean just running the ADC at 16bits doesn't really qualify as 'digitally converted' does it?

It would be interesting to know more about how they do it. (Edit: I asked the question in the SH forum so will see what they say)

Hi

If I spend the extra money to buy an ADC that really does a legit 24 bits at a 240 to 480 KHz sample rate, I've already put some money into the design. Bumping the FPGA up one notch to handle 24 bits rather than 16 bits is *way* less money than what I just paid to go from 16 to 24 bits on the ADC. It does not seem like a logical decision for a designer to make. It sounds a lot more like a marketing guy coming up with neat headlines. There is no reason to toss out 8 bits just to get to I/Q samples....

Some basic math:

16 bit ADC <$10
24 bit ADC >$40

16 bit capable FPGA ~$12
Same FPGA good to 24 bits ~$18

(yes, there are some assumptions in that math)

Bob

[hendorog]

Hi

If I spend the extra money to buy an ADC that really does a legit 24 bits at a 240 to 480 KHz sample rate, I've already put some money into the design. Bumping the FPGA up one notch to handle 24 bits rather than 16 bits is *way* less money than what I just paid to go from 16 to 24 bits on the ADC. It does not seem like a logical decision for a designer to make. It sounds a lot more like a marketing guy coming up with neat headlines. There is no reason to toss out 8 bits just to get to I/Q samples....

Some basic math:

16 bit ADC <$10
24 bit ADC >$40

16 bit capable FPGA ~$12
Same FPGA good to 24 bits ~$18

(yes, there are some assumptions in that math)

Bob

Well maybe you are right, you have a lot of knowledge and experience. However I can't help but think you are jumping to conclusions:

Firstly, it isn't in any marketing I have seen. The device has been out for years and this recent post about bit depth is in response to a question which was asked. The marketing says '16 bits'
Secondly, this is an inexpensive device. Its been well documented that they have built in some LO switching cleverness to remove spurious and keep the hardware price down. So why would they push the price up by adding a part that isn't required?

[uncle_bob]

Hi

If I spend the extra money to buy an ADC that really does a legit 24 bits at a 240 to 480 KHz sample rate, I've already put some money into the design. Bumping the FPGA up one notch to handle 24 bits rather than 16 bits is *way* less money than what I just paid to go from 16 to 24 bits on the ADC. It does not seem like a logical decision for a designer to make. It sounds a lot more like a marketing guy coming up with neat headlines. There is no reason to toss out 8 bits just to get to I/Q samples....

Some basic math:

16 bit ADC <$10
24 bit ADC >$40

16 bit capable FPGA ~$12
Same FPGA good to 24 bits ~$18

(yes, there are some assumptions in that math)

Bob

Well maybe you are right, you have a lot of knowledge and experience. However I can't help but think you are jumping to conclusions:

Firstly, it isn't in any marketing I have seen. The device has been out for years and this recent post about bit depth is in response to a question which was asked. The marketing says '16 bits'
Secondly, this is an inexpensive device. Its been well documented that they have built in some LO switching cleverness to remove spurious and keep the hardware price down. So why would they push the price up by adding a part that isn't required?

Hi

Exactly my point: Why would they spend a lot of money on an expensive 24 bit ADC (they are expensive) and then not use it? They could save a lot of money just buying the lower cost 16 bit instead.

Bob

[KE5FX]

The guy in question knows what he's doing, so he must've had a reason that isn't apparent from the outside looking in.  For instance, maybe they just stumbled on an unusually good deal on a bunch of those 24-bit converters on eBay.   In low-to-medium volume production with aggressive cost targets, part selection is as much an input to engineering as it is an output.

There could be purely technical reasons as well.  Antialiasing requirements are somewhat relaxed with sigma-delta converters since they oversample by nature, and that's a win because filters are always a hassle.  Especially if you're designing a spectrum analyzer, where you'd like to achieve spur rejection in the vicinity of 100 dBc or better.  So either he spends a lot of money buying ready-made LPFs from Mini-Circuits or whoever, or he spends a lot of board space (likely with accompanying shielding requirements) on discrete inductors and capacitors.  Unintuitive choices can make more sense when you're simultaneously fighting for space, power, and cost like they are.

[uncle_bob]

Hi

To be very clear, I'm not saying anybody is lying. I'm just saying that, on the face of it, it does not make any sense.

Bob

[hendorog]

Hi

To be very clear, I'm not saying anybody is lying. I'm just saying that, on the face of it, it does not make any sense.

Bob

Yep understood. Hopefully Justin replies to my question and we can find out more about it. Like KE5FX says there must be a good reason for it.

Maybe I'll even get some beer back

[bson]

The AD9864 appears to be a highly integrated digital IF subsystem and the ADC itself is just a part of it.  Perhaps it's simply not available in a 16-bit version?

[uncle_bob]

The AD9864 appears to be a highly integrated digital IF subsystem and the ADC itself is just a part of it.  Perhaps it's simply not available in a 16-bit version?

Hi

The sigma delta in the 9864 is exactly the sort of part that I was talking about. At reasonable data rates it is a 16 bit converter. As you decimate the output, you gain bits through the filtering process. At very low clock rates you fill up more of the i/o word with information (as opposed to noise). At the full rates they advertise for the Signal Hound, it's a 16 bit part. In very narrow bandwidths it is a bit better than that.

Simply put, sure, it's a 24 bit part in some cases. No it's not a 24 bit part the way they use it. The way they use it, it's a 16 bit part. Thus the 16 bits into the processing end of things. Makes perfect sense.

Bob