Market for an A/D converter with 150dB dynamic range?

Don't you believe in a tuned input? Even the simple tuned loop antenna in a cheap radio would reduce the level of the 710 and 780 KHz signals. I listen to WSM (50 KW) in Nashville, Tn. on 650 AM, and there is a local station in "The Villages" on 640 AM. As long as they are not over modulating, I can pick up WSM with no trouble. Try it on some crappy electronically tuned radio with no front end filtering, and you won't hear anything. I have used, and built HF and MW receivers for over 40 years. A crappy design is just that. You will still need AGC on DX signals to reduce fading, unless you want to ride an RF gain control, as well. I started with regen, then super regen receivers before I moved on to superhet, multiple tuned RF stages, then dual and triple conversion designs. The more work you do to filter the signal at the front end, the easier it is to process at later stages.

No, the noise will add up to 3dB louder, the signal will be 6dB, if it is correlated. So to get more dynamic range, put all the inputs in parallel and the noise improves by 3dB/doubling.

In digital land, the peak level is determined by spanning the maximum number of bits. However, the music sources are not correlated. Singers have to breathe, guitar notes hit a peak then decay, etc. So you mix signals that are not at their peaks at all times, but the noise is there as a constant signal. Hence the noise floor rises as you mix all these signals.

Others have pointed out that this is a fallacy; blending of any two uncorrelated sources of signal and noise would result in 1.4 volts of signal and 1.4 millivolts of noise.

Signals aren't just the voltage at one instant, they're averages of variations over long time periods; you have to do the addition of all the variations and redo the average on the sum to get the right answer.

Seriously, yes. The Geophysical market is the main user of the highest end ADCs available. 24bit ADCs used here have a usable dynamic range approaching 130dB with very low power consumption.

The best ADC designers in the world can't get better than 130dB or so for a few KHz bandwidth, but if you think you can do better than this you'll be a very rich man, start work on it now and make sure you invest all your time and money in it ;-)

Dave :)

David,

Thanks for pointing out seismic measurement as a market. Let me ask a question: while people are talking about the noise floor in this group and it seems we have reached to a physical limit of heat noise, why don't the seismic measurement devices have the same problem? How are they taking the advantage of 130dB dynamic range?

D.S.

Mainly because the dynamic range is specified over a lower bandwidth. Dynamic range will increase with reduced bandwidth. Have a look at the datasheet for this geophysical device for instance:

BTW, the typical specs on this type of device are a bit conservative, you can actually get better performance than that.

A few hundred Hz bandwidth might be useless for the audio industry, but for the Geophysical seismic market, bandwidths of only a few dozen or a few hundred Hz are all that is needed. Hence you geat greater dynamic range, and lower power consumption. Low power consumption is important because a typical seismic system will have thousands of these devices spread over many kilometers of cabling, either floating behind a boat in the ocean, on the sea bed, or on land.

Dave :)

Yes, but if you don't wish to overload under any circumstances you have to take to case where both inputs are at their maximum. In this case you get 2V of signal and 1.4 millivolts of noise.

kevin

In article , DigitalSignal wrote: [...]

This sounds like they have reinvented the instantaneous floating point converter. At low frequencies, 150dB would be no big problem to do with an IFP converter.

_Believe_ in a tuned input?? I'm not sure what that means. As it turns out, a filter in front of any active devices is not an option for me. That translates to needing high spurious free dynamic range. In any event, if you have a look at my posting to which you replied, you'll see that I DID explicitly mention filtering.

Interesting you think that filtering for signals spaced the same absolute frequency differences is easier at 9.8MHz than at 0.75MHz. The same attenuations will require loaded Qs 13 times as big. Especially given that the filtering should track the LO, and given the required loaded resonator Q (however many poles, and whatever shape you want in the filter), getting any significant attenuation about 1% offset from the center frequency should be an interesting design challenge. In addition, mechanical tuning is simply not an option in a great many applications. Maintaining frequency tracking among the LO and one or more resonators (connected to a source of unknown and varying impedance) to something noticably better than 1% over an octave bandwidth should present an interesting challenge.

Though the inductances are a bit easier at 10MHz than at 0.7-0.8MHz, I'd much rather deal with the far lower loaded Q that would be needed for the MW filter. But in any event, as I noted above, filters aren't an option for me anyway.

What in-band IIP3 will your receiver have? What degree of filtering attenuation will actually be required to achieve the performance goal I suggested, given that IIP3? Numbers will speak louder than generalizations.

Cheers, Tom "I have used, designed and built high dynamic range wideband inputs for spectral analysis applications for more than 40 hours."

You can't even use a bandpass filter to keep out of band crap from causing desense problems?

Why is the input uncontrolled impedance? A 3 dB pad will make it match better than a direct input, if you can stand the loss. What are the actual frequencies you're interest in?

Electronic tracking is possible, with varactor tuned RF stages. Either by creating a lookup table, or using an ADC to measure system gain, and a DAC to set the tuning voltage per stage. It was even done with vacuum tubes and servo motors turning butterfly capacitors in the '60s for the VOA Bethany, Ohio transmitters built by National Radio.

The last reviver design I worked on is still under an NDA, and I no longer have any of the paperwork. All I can say is that it was DSP based telemetry equipment.

All of the test equipment in my home electronics shop was badly damaged two years ago during the hurricane season when the roof sprung hundreds of leaks, so I can't make any measurements. The project has been on hold since then because I need $4000 for a new roof, and at least that much more to replace equipment that can't be repaired. I can't do the roof work myself due to my recent disabilities. I can't afford to have done, so everything is at a standstill, maybe forever.

It's common to have amplifiers with "noise figures" lower than 3dB. At

-174dBm/Hz. Amplifiers operating even at room temperature can have noise figures well below 3dB. How useful that really is depends on how noise-free the input signal is. For example, the signal from a resistive strain gage operating at room temperature won't be better than that -174dBm/Hz, but the noise signal from a dish antenna pointed at certain places in space and looking at signals in the GHz region could be very much lower. Practical amplifiers operating at room temperature can have noise figures that are a small fraction of a dB.

If I'd like to detect small signals--signals with perhaps -150dBm power in a 100Hz bandwidth--then I'd like to have instrumentation that has a noise figure under 3dB to get a decent signal-to-noise ratio. If I need to detect those signals in the presence of other large signals, perhaps as big as -20dBm or even larger, that sets the full-scale that the ADC must handle, because if the ADC is overloaded, all bets are off on being able to detect the small signal. Requirements like this set the allowable noise floor, of course, and also the allowable distortion. The small signals are resolvable from the large ones spectrally, but if I don't know ahead of time what frequencies either will be on, I can't use filtering to remove the large signals--the ADC just has to handle the whole range of amplitudes all at once. It's a tough problem to push down the noise and the distortion at the same time.

Such is often the life of people doing spectral analysis. There may well be other "axes" along which the resolution could take place, but in general, you're going to need digitization that captures a faithful representation of the input signal, covering a wide (enough) dynamic range to handle the largest inputs while keeping noise, distortion, and spurious signals low.

Cheers, Tom

that would be a 6db increase in power, which would reduce the signal to noise ration, as the noise only goes up 3db in power unless you have a phase matched noise source :)

Bye. Jasen

Not necessarily. If the two 1V audio sources are correlated (identical) then they will make 2 volts at the output. The noise in the two channels will not be correlated because it comes from two different sources so their powers will add and their combined amplitude will only be root 2 times their original values. The signal/noise ratio will now be 3dB greater. That's why a magnetic tape head covering twice the tape width has a 3dB better S/N ratio.

Ian

Ah, isnt the Internet wonderful. One guy says mixing sources increases the noise. I suggest common sense thought experiment suggests the noise doesnt get worse. Then some really smarty-pants says due to the non-correlation the noise goes down.

All these options... wunerful...

Are you really that stupid? The audio signals I was talking about are not in phase, or do you even want them all at the same level but the noise added by each channel is still there, and the more channels on the mixing buss, the higher the noise floor. If the signals were identical and in phase you would only need one channel, no mixing buss, and you would only have the noise from one channel.

Sigh. One last try. Try this thought experiment: You have ONE BILLION signal generators, all running off separate batteries, each putting out one volt, each with a microvolt of noise. The noise is

Now you hire 1,000,000 s, give them all some wire, and tell them to hook all the signal generators in series.

You take the two free ends and hook your "Harbor Freight $2.49 multimeter across the wires. Oops, meter blows out. There goes $2.49!

You go to your junkbox and pull out the first two resistors you grab. Lucky you, one is a 999,999,999 ohm 0.00000000000001% resistor. The other one is a 1 ohm 0.00000000001% ohm resistor.

You hook up the 999... ohm one to the hot wire, the other end of that one to one end of the 1 ohm resistor, and the other end of the 1 ohmer to the grond wire from the generators. So you have a ONE BILLION TO ONE voltage divider. Okay so far?

You go buy another harbor freight $2.49 meter. Put it across the one ohm resistor and what do we measure?

By your reckoning, we should measure, lessee start with -120db, add three db for each generator, that's, hmm, THREE BILLION DB. that's a lot of noise! Divide that by the voltage divider of a billion 10^9 in voltage, 10^18 in power, 180DB HMMM, we have 2,999,720DB OF NOISE!

By my reckoning, you'll measure one volt of signal, one microvolt of noise.

Now please explain where I went wrong and why your answer is right.

Where do you get your mytical noiseless 999,999,999 resistors? ?

we don't need any resistors, we used just two, only because our $2.49 meter doesnt go up to a billion volts.

And they don't have to be noiseless, if anything that would add a few microvolts to both our scenarios.