ADC overload myths debunked

Steve-N5AC

I've received some feedback that there is some confusion circulating on other ham radio reflectors regarding how analog to digital converters (ADCs) work in radio applications.  Specifically, some of the comments tend to say that direct sampling ADCs just won't work in strong signal environments so I'd like to explain why this is not factual for those who are interested. I have a few points to illustrate this.

As hams we tend to think of strong signals in terms of their total power, how many total Watts they are.  When you think of signals in this way, you can add their power in your head and think: two -10dBm signals add to -7dBm total power (3dB increase).  In fact, you can take multiple signals and add them together in a power meter and the power meter will show the total power of all signals.  But this is the average and not instantaneous power.

An ADC, on the other hand, is really a discrete signal device.  All of the signals get chopped into samples and so the real question is: how do the signals add together in the discrete time domain?  To answer this, we have to look at the signals and how they interact.  An RF carrier is like any AC signal -- it is a sine wave that varies from negative to positive voltage along the curve of a sine wave.  If we add two sine waves of exactly the same amplitude, frequency and phase, the peak voltage will be doubled (6 dB).

But two signals of the same amplitude and phase on the same frequency isn't reality.  Reality is signals all across the bands that are totally unrelated (uncorrelated) -- for example one at 14.100374 and another at 21.102392, etc.  The variance of the algebraic sum of these signals will decrease with the square root of the number of signals present.  As more signals are added, there is a decreasingly small probability that these signals will add (precise alignment of the highest voltage peak of the signals) and the algebraic sum of the signals will degenerate into a quasi-Gaussian distribution.  To get a fabled 6dB voltage rise, they would have to already be exactly the same voltage, frequency and phase (this is what is done in a power combiner in an amplifier and it’s hard to make that happen).  If one is stronger, the addition of a weaker signal will not add much to the total level.  

If we're talking about a large number of signals across a wide spectrum, it's the same situation.  They would virtually never all add at the same time so they will not combine at just the point where the peak of all signals occurs.  It just doesn't ever happen.  As a mathematician friend of mine pointed out, the two primary principles involved are the Law of Large Numbers (https://en.wikipedia.org/wiki/Law_of_large_numbers) and the Central Limit Theorem (https://en.wikipedia.org/wiki/Central_limit_theorem) which you can peruse for more insight.

As an intuitive analogy, we could look at our solar system.  Let's discuss the likelihood that the planets will cause the ocean to rise and cover up the state of Hawai'i. The planets all have their own period around the sun (frequency).  They are all different amplitudes as well (gravitational influence on the Earth if we're thinking about rising tides).  The questions are:

1) How often do all the planets align?
2) When they do align, will the ocean cover Hawai'i (overload)

There was a book published on this in the 70's called The Jupiter Effect (https://en.wikipedia.org/wiki/The_Jupiter_Effect) which proclaimed death and destruction when this was to occur.  The book was, of course, proved wrong but not before it became a bestseller.  First, the planets almost never come into alignment -- even in the book the planets were only going to be on the same side of the sun, within a 95-degree arc.  Second, when they do align, the amplitude from the outer planets is so low, it just doesn't matter.  My college physics professor was asked about this problem and worked the equations and showed that even if they were all in precise alignment, the ocean would rise by an additional 1/4" briefly... just not worth worrying about.  It is the same situation in ADCs.  The real truth is that more and stronger signals actually make an ADC work better through a process called linearization.  Everyone that has studied ADCs knows this -- the irony here is that lots of strong signals are a benefit, not a detractor like they are in old technology superheterodyne transceivers where IMD dynamic range degrades rapidly with signal strength.  Translation: Strong signals -- Bring it!

Another point to make is that all overloads are not created equal.  Overload sounds like an undesirable situation, but a momentary overload has no significant effect on a direct sampling radio.  Why is this so?  The individual data points that make up a signal you are listening to are almost never going to fall in the same time as the overload, statistically.  With a noise blanker, we remove thousands of samples with no negative effects to the signal being monitored and a momentary overload from the addition of many signals summing up will have a much lower effect.  This effect is called "soft overload" because momentary overloads just don't have an impact on the radio.  It takes much more significant and sustained overloads to cause a real problem.  The overload that folks are talking about is a non-event.  Even if it did happen, it's not going to affect the radio's performance.

Finally, there's often confusion about dynamic range from wideband ADCs.  The confusion generally works like this -- someone will lookup a data converter that runs at 100MHz and see that it has a dynamic range of 70dB and assume that it could never beat a radio with an 85dB dynamic range.  The problem is that this is an apples and oranges comparison.  You cannot talk about instantaneous dynamic range without talking about detection bandwidth.  For ham radio, this is the width of the actual receiver.  We use a standard 500Hz bandwidth receiver for comparison purposes but it could be 2700Hz for sideband or 50Hz for CW, for example.  

What really happens is that we use a process called decimation ( https://en.wikipedia.org/wiki/Decimation_(signal_processing) ) which takes the data collected at an oversampled rate (100MHz for example) and then systematically reduce the sampling rate down to the bandwidth of interest.  In this process dynamic range is increased in what is called "processing gain" (http://www.dsprelated.com/freebooks/sasp/Processing_Gain.html).  In the FLEX-6500 and FLEX-6700, we operate the ADCs at 245.76 Msps so that the typical processing gain is on the order of 56dB.  When added to the 75.5dB quoted spec of the ADC, the calculated instantaneous dynamic range is on the order of 132dB.  This far exceeds the dynamic range of ALL superheterodyne receivers (Don’t believe what you read about blocking dynamic range as it is irrelevant if the radio falls apart due to phase noise before this level).

In reality, it is impossible for any receiver to have blocking dynamic range or IMD dynamic range greater than its phase noise dynamic range (PNDR) otherwise known as reciprocal mixing dynamic range (RMDR).  In all cases and no matter the architecture, if RMDR is less than BDR or IMD DR for a given tone spacing, the phase noise will cover the signal of interest before blocking or IMD will be a factor.  In fact there is not a single transceiver from any manufacturer on the market that would not have its blocking dynamic range limited by its internal phase noise much less first by the noise from the transmitted signal.  

Most of the old technology superheterodyne transceivers on the market have horrible RMDR numbers.  When a strong signal is heard by them, their oscillators spread the signal all around the band as noise covering up signals you are trying to hear.  Here's the simple test: Take two of your favorite legacy radios and transmit in one while listening in the other and watch what happens to the noise floor at 2, 10, 20, 50 and 100kHz from that signal.  You will see that these receivers show significant noise floor increases that prevent operation near each other.  This is the practical concern -- there's no reason to talk about a number of mythical strong signals of all the same power that might correlate to cause an overload in a new type of receiver... the real problem is the superheterodyne receiver that folds under a single strong signal in the vicinity of small signals you are trying to copy.  Most contesters have experienced this first hand when two radios are being used.  If you have to tell your operating buddy in the same band to stay so many kHz away from you, you know the problem well.  This is also a classic Field Day problem.

We have thousands of radios in the field and if any of these issues were real, we (and you) would have heard about it.  You should have confidence that you have the best transceiver on the market -- experienced and knowledgeable people have said so.  They have said so because it is proven out in test after test and it is simply mathematically true.  FlexRadio Systems makes the best amateur transceivers available.

DrTeeth

Thank you for another article clarifying an otherwise difficult topic to this (almost) OM.

It is a rare talent that you have. Keep 'em coming.

VK7WH Winston

Well said. I second that Winston VK7Wh

W9OY

I have confidence as well as experience  

73  W9OY

Bob G W1GLV

You cannot beat an engineer that knows what he is talking about. Fine job Steve.

Corey/ KC0YNS

Thanks Steve,  Spoken like a true engineer, Over my head, But I'm learning....

Bill -VA3WTB

I just love it when he does that!!!

Bob-N4HY

Steve it is awesome, just awesome to have watched the growth of your understanding and the never ending quest for knowledge that drives you every day of your life.  I want to know when you want to come and guest lecture my communications theory students.   Spock to Kirk: "You have been and always shall be my friend".

Greg Zenger [N2GZ]

QED

Barry N1EU

Thanks Steve, great post.  My feeling all along is that the guys on the E reflector are just missing the boat.  The days are numbered for superheterodyne front-ended flagship hf xcvrs.

Steve Conklin

Thanks, Steve !

Please make this available as a paper or pdf posted somewhere outside the community site, as I don't think that the people who could most benefit from this are members here.

Steve-N5AC

Thanks, Bob.  You made my day whole week.

Peter K1PGV

Super interesting. The ability to clearly describe complex technical concepts this well is rare. Well done, Steve. Peter K1PGV

DrTeeth

We need to have a "Steve's Corner" for all of his great posts. Then, when there are enough, they should be (self-publushed?) as a book. I'd be the first in the queue to buy it.

Mike W8MM

Jupiter and Hawaii was a great analogy to vivify the explanation.  Great writing!!!!

Peter Bentley

 FlexRadio Systems makes the best amateur transceivers available.

Agreed..... Thank you Steve

KD4HSO

Great write-up.  I take it, from the 75.5 dBFS SNR, you guys are using the AD9467-250?  Can you share what clock synthesizer you are using?

Steve-N5AC

Yes.  It's not a synthesizer -- synthesizers have too much phase noise.  It is a disciplined crystal oscillator running at 983.04MHz (FLEX-6700 and FLEX-6500)

EA4GLI

Hi Steve. Great article. I took the liberty to translate it into Spanish and publish it in the URE.es forum giving you credit as author. I hope you are ok with it. Otherwise I can promptly remove it.

Gerald-K5SDR

Yes, are using the AD9467-250.  The 75.5 dBFS is specified at the Nyquist bandwidth of 122.88 MHz based on the sampling rate of 245.76 Msps.

Gerald-K5SDR

No problem on posting the translation.

Tim - W4TME

This technology description is also available as a link on our HelpDesk
https://helpdesk.flexradio.com/hc/en-us/articles/205527943-ADC-Overload-with-Direct-Sampling-SDRs-My...

KD4HSO

What's the ADC clock jitter?  Sorry to pester you guys with question; just curious as to what is achievable for amateur use given price/performance.  Had my 6500 for 1 week and loving it.

Marc Lalonde

HI  it   VF901268  from CTS Valpey-Fisher

good luck for find any info on that part ,seem semi custom or at least hard to find 

Steve-N5AC

It is a custom part and jitter is not really the right way to look at a sampling clock.  Jitter is integrated phase noise and we really want specific numbers at specific offsets.  The phase noise of the clock is approximately -98dBc/Hz @ 100Hz, -127dBc/Hz @ 1kHz, -147dBc/Hz @ 10kHz, -153dBc/Hz @ 100kHz, -156dBc/Hz @ 1MHz

EA4GLI

Do you prefer if I link to the helpdesk url instead of this forum post? Thanks Gerald.

Tim - W4TME

Salvador - either way will be fine. 

EA4GLI

OK Thanks Tim. FYI here is the post on ure.es http://www.ure.es/foro/6-tecnico/223735-todo-flex-y-smartsdr.html?start=60#291760

I try to keep the spanish hams abreast of Flex updates on that forum. URE is Spain ́s ARRL. Not all of the EA understand English and Google translate butchers technical texts. For example, it translates Hams as the cured meat and not as amateur radio operators.

Tim - W4TME

Muy bien mi amigo. Hams no son barbacoa!

KD4HSO

I thought jitter set the ENOB (effective number of bits) from the ADC, which I assume is around 12, which limits ADC dynamic range. So for closely spaced tones, which is main concern in this application, does the phase noise mask directly image onto the received tones, just like a standard superhet?

Steve-N5AC

In a well designed synthesized superheterodyne transceiver, the low noise master clock will generally be the highest frequency where you can acquire good phase noise.  Then the this clock is divided down in either a DDS or with a PLL (the former has significantly better phase noise properties).  As you go lower in frequency, the division effect in the synthesizer produces better phase noise -- it is not generally consistent across the served frequency range.

In a direct sampling receiver, the phase noise at the sampling clock is turned into phase noise in the receiver at the sampling frequency.  This means that the phase noise present at the given offsets is not divided down for better effect.  This makes the phase noise requirement on a direct sampling clock much higher than that for a synthesizer.

Jitter is integrated or averaged phase noise.  It does not tell you, specifically, at what offsets you are most likely to observe the clock at any given time.  The type of source clock will have a phase noise pattern based on how the clock functions, if it locked to a source, the loop bandwidth on any PLL locking the clock, etc.  The detailed phase noise allows us to calculate noise floor rises at offsets from carriers that just knowing the jitter would not allow us to do.  It also provides information on the application that is useful.  

In HF radio applications, phase noise to about 100kHz is important because we have strong signals in that range, especially in FD and M/x applications.  In microwave (say 10.3681 GHz operation) though, we are significantly more concerned with close-in phase noise.  Because clocks are multiplied up, a small offset (10Hz) may end up being 1kHz after multiplication and phase noise here could interfere with a CW single close to you.  On HF, 10Hz phase noise isn't generally as important.