Re: Softrock Dynamic Range and MDS Redux
- Thank you Milt and Tony! A wonderful explanation of the phenomenon I have been observing.
My follow-on question would be: Could we incorporate the spectral noise level on the FFT display into the calculations of MDS and dynamic range and, if so, how?
Warren Allgyer - W8TOD
--- In email@example.com, Milt Cram <w8nue@...> wrote:
> On 2/18/2013 2:55 PM, Anthony Casorso wrote:
> > I hope that this is not redundant since I have not read all of the
> > posts in this thread. The conventions that have evolved for defining
> > SNR, MDS etc have all come from a history where the ultimate receiver
> > of the signal is someone's ear. In that context, the noise and it's
> > relationship to the bandwidth make sense because the ear doesn't have
> > the ability to ignore noise at one frequency while paying attention to
> > noise at another frequency. The ear hears all of the noise in the
> > passband of the filter and it obscures the intended signal even if it
> > is not at the same frequency as the intended signal. When you switch
> > to a world of FFTs and "visual reception", the story changes. When
> > the eye is the receiver, the presentation of the signal and noise
> > allow for the possibility of ignoring noise that is in the filter
> > bandwidth but not at the same frequency as the signal. The effect of
> > ignoring the noise off to the sides is to narrow the filter bandwidth
> > to the width of the signal. In an FFT display, this is the FFT
> > "equivalent noise bandwidth". In an NDB receiver project that I have
> > been working on that used SpectrumLab, I defined two SNR's. One I
> > called the Visual SNR and the other the Aural SNR. Aural SNR is the
> > traditional SNR where the signal is ratioed with the noise in the
> > specified filter bandwidth. It improves as the filter narrows and is
> > unaffected by the FFT bin width. Visual SNR is the ratio of the signal
> > to the noise with a bandwidth equal to the FFT bin equavelent
> > bandwidth. This is what the eye sees on an FFT display.
> > Tony
> > ------------------------------------------------------------------------
> > To: firstname.lastname@example.org
> > From: allgyer@...
> > Date: Mon, 18 Feb 2013 16:01:51 +0000
> > Subject: [softrock40] Re: Softrock Dynamic Range and MDS Redux
> > Completely agree.... the RMS value as measured on a voltmeter at the
> > speaker terminals is the same. However, if you look at the two signals
> > superimposed on a spectrum display, the noise power is dispersed
> > across the bandwidth of the filter. The signal power is concentrated
> > in a single "spike" and, therefore, measures about 20 dB higher at
> > it's peak.
> > The total power of the noise and the signal are the same. The power of
> > the signal generator is focused on a very narrow bandwidth and,
> > therefore, appears to be much greater on a spectrum display.
> > Warren Allgyer - W8TOD
> > --- In email@example.com
> > <mailto:softrock40%40yahoogroups.com>, "g3vnc" wrote:
> > >
> > > --- In firstname.lastname@example.org
> > <mailto:softrock40%40yahoogroups.com>, "warrenallgyer" wrote:
> > >
> > > > - This signal is called "MDS" even though it is quite high, 10 -
> > 100 times the average voltage of the noise itself and easily audible
> > on the receiver.
> > >
> > > I don't understand this statement.
> > >
> > > Although the MDS and noise are spectrally quite different, if the
> > MDS power and the noise power in the resolution bandwidth are the same
> > e.g. -150dBm, then the RMS value of the MDS and the noise in that
> > bandwidth are also the same.
> > >
> Hi all,
> In all of the discussion thus far, I have heard no mention of "power
> spectral density". In SDR applications, where an FFT is usually used to
> produce a spectral display, we see such a power spectral density. That
> is, the display indicates the power versus frequency. When using the
> FFT, the minimum frequency resolution is the "bin width". If the scale
> of the spectral display is calibrated in watts, then each bin represents
> a certain number of watts divided by the bin width in Hz. If one then
> introduces the bandwidth of the particular operating mode, the total
> noise power is the integral (sum) of the power spectral density over the
> bandwidth--e.g. the observed power in an FFT bin times the number of
> bins within the bandwidth. For CW, where the signal may be contained
> within a single bin, and noise is distributed over many bins, equal
> noise and signal power means that the noise level on the spectral
> display could appear to be much less than the signal. However, if one
> was operating FM, with the signal smeared over the entire operating
> bandwidth, the signal would be difficult to separate from the
> noise--even with the same SNR as in the case of CW. PSK 31 is another
> example of where the signal energy(power) may be concentrated in one or
> two bins of the FFT spectral display.
> If the length of the FFT is long, the bin width can be quite small--i.e.
> a few Hz. If the length of the FFT is short, the bin width may be
> measured in kHz.
> In addition, it should now be clearly evident why one is able to "copy"
> a CW signal using a narrow band filter, when it may be difficult, or
> impossible, when using a wide bandwidth filter. If the filter bandwidth
> nears the bin width, the SNR will be significantly improved.
> In summary--don't be confused by the spectral display when considering
> SNR, MDS, and dynamic range.
> My 2 cents--
- Good thought Victor.
Crosstalk between the generators is a good possibility since they are on the same Silicon Labs 5338 EVB. So, of course, I set up a test.
Generator A was set for 7.050 and Gen B for 7.055 and both fed through the hybrid, then through a total of 31 dB of fixed/variable pad to the Softrocks.
Alternately disconnecting Gen A and B I measured B= -26 dBm and A= -25 dBm. With A disconnected I measured -116 dBm on 7.050 and with B disconnected I measured -98 dBm on 7.055. So A crosstalks into B at -90 dB and B into A at -72 dB. Not great but at least I know what I have.
I do not have a second variable attenuator so I put fixed 20 dB pads on the output of both generators and reduced the variable/fixed combo after the hybrid to 11 dB. The readings were identical. So the crosstalk is on the board and not in the hybrid.
I also checked the IMD3 products at 7.045 and 7.060. The levels were very comparable in both cases, down about 70 dB from the main signals. In both cases the IMD3 products stepped linearly with the main signals until I reached an input level of -25 dBm which is the op amp overload point and they skyrocketed.
What does this prove? Maybe something, maybe not. It is possible the IMD3 products generated by the onboard crosstalk are so high they mask the true IMD3 generated in the radio.
In order for that to happen the "real" IMD3 product would have to be at -96 dBm or less when the main signal is at -26. If it is at -96 and hidden then that would make the intercept point at +44 dBm, an almost unheard-of number. But not impossible.
Somewhere in these posts there is a reference to tests done on the FST3253 mixer that quoted test numbers in this range.
My bottom line on this is that the IMD3 performance of the Softrocks receiver, at least prior to the op amps, is so good that it is irrelevant.... or that the third order intercept test itself is irrelevant for this breed of receiver, as is maintained by the ARRL handbook.
As usual, probably a lot more information than anyone really wants to know. But you gave me a real fun morning Victor!
Warren Allgyer - W8TOD
--- In email@example.com, "victor" <victorkoren@...> wrote:
> Warren, I bet that your two generators make intermodulation between them so they generate the distortion signals, and that's the reason that the intermod level stays the same (compared to the generators signal level) when you change the level by using a variable attenuator at the receiver input. If you have an additional variable attenuator, do the test again but now put the variable attenuators in series with each generator output before the hybrid coupler and change them together to change the signal level at the receiver input. You will see that at lower signals (higher attenuation) the distortion will go down faster because the isolation between the generators will increase.
> Victor - 4Z4ME
> --- In firstname.lastname@example.org, "warrenallgyer" <allgyer@> wrote:
> > Nick (and others who contributed to this very interesting discussion):
> > I finally got back to my workbench and decided to tear into this issue again. I set a two channel signal generator (homebrew Si5338 referenced to GPS) to generate two signals, 5 kHz apart, at about -30 dBm through a hybrid coupler and a step attenuator. The third order products were readily visible on the baseline about 55 dB down.
> > I tried to confirm the 3X principle where the IMD3 products increase at 3 times the rate of the fundamental but, in my case, they did not. A 3 dB increase in the fundamentals resulted in about a 3dB increase in the IMD3 products as well. It seems to be linear, not 3X!
> > Back to the handbook to check my methodology and I come upon this little nugget on Page 25.32 in the "Receiver Dynamic Range" measurement section:
> > "The third-order intercept is generally not a valid concept for software-defined receivers (SDRs) that do not use an analog front end. Some SDRs do not use a mixer but feed the signal from the antenna directly to an analog- to-digital converter (ADC). ADCs usually do not exhibit the 3 dB per dB relationship between signal level and third-order products, at least over major portions of their operating range. Comparing third-order dynamic range measurements of an SDR and a conventional analog radio may give misleading results."
> > SO...... it seems that we cannot use the third order intercept principle to establish the upper number of the dynamic range for the Softrocks.
> > I am at a bit of a loss. It seems that the practical upper number for the Softrocks seems to be the level at which the op amps start to clip which seems to be in the -12 to -20 dBm range, and that 3rd order intercept is a fairly useless number in these cases.
> > Any thoughts?
> > Warren Allgyer - W8TOD
> > > > On the two-tone intercept: I think I understand the methodology. IMD3 products increase at 3x the rate of the two input signals. So if you measure the product, remove 10 dB of attenuation, measure again... you get two points on a line.
> > >
> > > Actually you only need one point because you know the slope of the line. But two (or more) improves the accuracy of the measurement.
> > >
> > > +30dBm is where my measured 3rd order line meets the 1st order line on my graph.
> > >
> > > > Do you, or anyone, understand the rationale for designating 2/3 of the resulting range as the maximum signal level? Is it just arbitrary or is there some very quantifiable effect on the reception of desired signals at that level?
> > > >
> > > It's not an arbitrary designation. The overload level is defined as that level where the 3rd order products are at the same amplitude as the MDS. It follows that the Two Tone Dynamic Range is 2*(Pi-MDS)/3. There is an algebraic explanation in one of the Appendices to "Solid State Design for the Radio Amateur". It can also be derived graphically.
> > >
> > > > As a number that is replicable and comparable between receivers I can see its' value. I would like to understand though what a signal 100 dB above MDS would sound like compared to 96 dB.
> > >
> > > In this case with one such signal it would sound 4dB louder if you switched the AGC off. But with two such signals the 3rd order products would now be 12dB above the Noise Floor (3 x 4dB) i.e. at -103dBm in my Softrock at 5MHz. Would you hear this? Quite possibly, if the two signals were outside the passband and one of the third order products fell within the passband.
> > >
> > > HTH
> > >
> > > 73 Nick G3VNC
> > >