Multipath Simulation for SSB FM Stereo

[gtomas]

Clip here: http://www.cgould.com/sound-files/SonyLRDemod.wav has both channels with the right channel pulled.

I've never listened to XDR-F1HD myself - apparently they don't make them for European market that has no HD Radio and with 50us de-emphasis...

But the artifacts are quite obvious in the clip above, in the dead channel. They do indeed sound swishy, swooshy MP3 like. Sometimes even R2D2 like

I'd too be interested in the full stereo sample, to hear if that's something that's noticeable when masked with music. It sounds like it might be, but masking is quite perceptually powerful so...


Regards,
Goran Tomas

 

[k6sti]

I've updated my multipath simulation once again.

1. I increased the FFT size to 16384 samples. This yields a frequency resolution of 61 Hz. Many intermod products resolved into multiple peaks at the higher resolution. I may have overlooked some, but I did a quick check and found no harmonic products, only intermods. I wonder if this is what gives multipath distortion its characteristic sound.

2. I was curious what happened for very long multipath delays. Since the SSB S/N advantage doesn't change much with multipath level, I figured a single graph probably could tell the story. I added one that plots the figure for a multipath level of -12 dB from 1 µs to 400 µs in 1 µs steps.

3. I added a sound sample for real speech.

Brian

 

[FFoti1]

I've updated my multipath simulation once again.

1. I increased the FFT size to 16384 samples. This yields a frequency resolution of 61 Hz. Many intermod products resolved into multiple peaks at the higher resolution. I may have overlooked some, but I did a quick check and found no harmonic products, only intermods. I wonder if this is what gives multipath distortion its characteristic sound.

2. I was curious what happened for very long multipath delays. Since the SSB S/N advantage doesn't change much with multipath level, I figured a single graph probably could tell the story. I added one that plots the figure for a multipath level of -12 dB from 1 µs to 400 µs in 1 µs steps.

3. I added a sound sample for real speech.

Brian

Brian,

What's your motive here?

-Frank Foti

 

[k6sti]

Brian,

What's your motive here?

-Frank Foti

To make the simulation yield greater detail, to provide more comprehensive results, and to add another dimension to its output.

Brian

 

[k6sti]

I've just added two different IF filters to the multipath model. It's interesting how the SSB S/N advantage numbers changed in the delay vs attenuation table. I added graphs for SSB and DSB distortion due to the IF filters (no multipath). Finally, I added a graph for the RF/IF signal spectrum for mono, SSB, and DSB so you can see how much narrower the SSB signal is.

http://ham-radio.com/k6sti/ssb.htm

Brian

 

[k6sti]

I've updated the multipath simulation as follows:

1. As a model check, I tabulate the monophonic S/N advantage over DSB for a wide range of multipath delays and levels. See if you can guess the average value of the numbers based on your personal experience before looking at the results.

2. I've updated the IF filter model for something like the fourth time. I was able to get both the amplitude and group delay of the IIR model to be essentially identical to those of a cascaded pair of 280-kHz Murata ceramic filters, which is a common FM IF filter. The exact response turns out not to be critical, but it's nice to have an accurate model. The IF filter does affect the SSB S/N advantage.

3. I graph total harmonic distortion due to two IF filters (equivalent to two and four ceramics), including the effect of deemphasis. DSB and SSB yield somewhat different distortion values.

4. I plot curves for out-of-band power for a wideband region and for the region HD Radio sidebands occupy.

5. I plot curves for DSB and SSB stereo separation for wide and narrow IF filters based on measured composite amplitude and phase values. I include a curve for the composite magnitude so you can see the problematic (for SSB) slope.

http://ham-radio.com/k6sti/ssb.htm

Brian

 

[FFoti1]

It will be interesting to see if these simulated results are the same, once they are tested using a complete transmission path, and broadcast engineering practices.

-Frank Foti

 

[Schroedingers Cat]

4. Has not caused any negative feedback regarding stereo separation.

An Oldies station would be the ideal test for that. Modern pop music has so little stereo separation that listeners would never be able to hear the difference anyway.

A prime reason why I'm using a lot of older Beatles songs for this. Perfect content to demo stereo separation.

-Frank Foti

The Beatles recordings and others from the 60s are really not true stereo though. There were heavily produced tracks that were put in one channel or the other with vocals usually in the center "channel". On the other hand, when recording an orchestra, there are two microphones that are nearly the same distance from each musician. Rarely would there be an instrument that was more than 6 dB different in intensity in the other channel. It is the audio phase shifts and the slight intensity differences that create the true stereo ambience. Those 60s recordings are fun to listen to though, especially with headphones, or by somehow producing an R-L effect, with an expander, vocal eliminator, or simple phase reversal.

 

[FFoti1]

The Beatles recordings and others from the 60s are really not true stereo though. There were heavily produced tracks that were put in one channel or the other with vocals usually in the center "channel". On the other hand, when recording an orchestra, there are two microphones that are nearly the same distance from each musician. Rarely would there be an instrument that was more than 6 dB different in intensity in the other channel. It is the audio phase shifts and the slight intensity differences that create the true stereo ambience. Those 60s recordings are fun to listen to though, especially with headphones, or by somehow producing an R-L effect, with an expander, vocal eliminator, or simple phase reversal.

The Beatles stereo tracks are good test examples that easily demonstrate the ability of a system to replicate the channel designations from the source recording. Considering that most of those older Beatles stereo tracks put the vocals in one channel, and the instruments in the other, this offers a simple audible method to hear if separation, in a system is compromised or not. This is the reason for using this content in testing the DSB/SSB systems for audible differences in separation.

-Frank Foti

 

[Bojcha]

Isn't just simpler to mute left (or right) source channel and compare ?

 

[Kmagrill]

"True stereo" or binaural audio, seldom has anywhere near as much perceived stereo seperation as is found in nature. The techniques used since stereo records came into vogue have permitted hifi and stereo radio users to hear a much more expanded and open sound than they would have heard using the two microphone techniques, in most cases. I have heard of binaural recordings that closely mimic the effect of sitting center stage during a musical event, but I have never heard one that actually seemed to compare to the live presentation and live presentations are usually pretty different from the studio versions. Generally, I don't think these types of recordings would be a very good test for a stereo transmission system since they don't tend to stress the l-r content very much.

 

[FFoti1]

Isn't just simpler to mute left (or right) source channel and compare ?

Actually, this is what we do to test the system with content. When the desire is not to mute one channel, then using content with extreme channel placement offers an alternative method.

-Frank Foti

 

[k6sti]

Latest multipath simulation updates:

1. I increased the dynamic range of the FM detector by 2.5 dB. This affected results only when the delayed signal was within 3 dB of the desired signal. Dynamic range and sample rate can be an issue for a DSP detector since co-channel interference and multipath can generate large instantaneous-frequency excursions and high bandwidth at the detector. Evidently the arctangent detector and 1-MHz sample rate I'm using are adequate. I measured the capture ratio as 1 dB. For comparison, I measured 1.1 dB for the Sony XDR-F1HD and 1.4 dB for the Sangean HDT-1, both of which use a DSP FM detector.

2. I added a high-frequency two-tone test signal consisting of 7.08- and 9.77-kHz tones. This complements the low tones at 732 and 1953 Hz. The two tone sets scatter power across the RF/IF spectrum in different patterns.

3. I refined the IF filter model yet again. I added measured curves of amplitude and group delay for two types of Murata 280-kHz ceramic filter so you can compare with the IIR model curves. For the previous filter model, the group delay was a bit high within the 200-kHz passband where most of the FM power winds up. Distortion values dropped considerably, although the spread between SSB and DSB distortion widened. SSB S/N advantage doesn't seem particularly sensitive to the IF filter although some differences can be seen.

4. I added, consolidated, and overlaid several long-delay graphs, mostly so effects of the two IF filter models could be more easily compared. I added long-delay graphs for a delayed signal just 1 dB down. This can cause a severe amount of multipath in stereo. I was concerned that the -12 dB I had been using, which is rather tame by comparison, might not reveal an advantage for one method or the other that arose only when things got really bad. Incidentally, the graphs go out to 400 µs, which is a path difference of 120km. That limit seemed rather high. I was going to shorten it until I made some measurements on a map and realized that I regularly experience strong multipath over a 75km path difference from a mountain range on the horizon.

5. I redid the speech sample at a more unfavorable delay value. I also left the 2.35 dB increase in audio level the multipath caused since that's what a listener will hear.

6. I added row and column averages to the tabulated numerical results to try to make the mass of data a bit more intelligible.

Brian

 

[WNTIRadio]

I've been following this thread Brian, and I do have a question:

Why are you just using tones? We don't listen to tones.

It's the same reason Frank, Bob, Cornelius and Leif will tell you not to attempt to set up your audio processor with tones, but with program material. Tones tell me that it'll be some weird "only in this way will this happen" result to your tests. When will someone be driving along listening to a two tone test?

What about just transmitting L-R for a test and listening to the result? All the crap is in there anyway, so a pure L-R signal with program material will tell you a lot of what's going on. Personally, on my stations I could give a rat's flying backside if the S/N dropped by 1 or 2dB, which will still put the total S/N way over any consumer radio, if the multipath is lessened.

But I do understand scientific processes, and the need for such. However, from all my readings it seems your simulation falls flat because it only assumes one source of multipath. Sitting in an urban canyon, there could be 15 different reflections reaching the radio at different times.

 

[k6sti]

> Why are you just using tones? We don't listen to tones.


I use tones for the same reason they have always been used in audio. Tones let me quantify things. They are repeatable and measurable in a way that program material isn't. And they can provide insight not possible with complex input. But I did run speech through the model as a check. Didn't you listen to the sound sample?


> Tones tell me that it'll be some weird "only in this way will this happen" result to your tests. When will someone be driving along listening to a two tone test?


I think you misunderstand the purpose of the model. It is intended to give insight into what's happening and to make controlled comparisons. It's not intended to provide a subjective comparison.


> What about just transmitting L-R for a test and listening to the result?


I think you need to listen to the sound sample I provided. It uses a single stereo channel, not L-R. I avoid L-R because it is not representative of typical program material and has no power at all at baseband, which limits the intermodulation possibilities. But listening isn't enough. I want to measure and quantify differences.


> But I do understand scientific processes, and the need for such. However, from all my readings it seems your simulation falls flat because it only assumes one source of multipath. Sitting in an urban canyon, there could be 15 different reflections reaching the radio at different times.


Didn't you see the plot I generated that contained five reflections? I can model any number, but I usually stick with one because it makes it easier to understand what's happening. Many reflections only add confusion. They are good to check but not so good for analysis.

I think you need to actually read the writeup and listen to the sound sample. They're here:

http://ham-radio.com/k6sti/ssb.htm

Brian

 

[FFoti1]

This is a prime reason why various simulations, done during algorithm development, do not play out in the real world. For years, I've done simulations on processed signals during the development stage. There have been times, where I got so deep into the development, using simulations, my thoughts truly believed the algorithm must work great because there was amazing peak control, and pure looking spectrum.

Sadly, many of those designs, fell flat on their face, when audio was applied, as we need to use it in the real world. Yet, the simulations looked incredible. I have also experienced the opposite, where a concept didn't look so good in simulation, but its performance was mind-blowing.

The same holds true here with SSB, and multipath performance. Simulations are a good starting point, and barely half the story. I have already run many of Brian's tests, using a complete broadcast transmission chain, and the results are significantly different than what he keeps telling us. Upon completion of our testing, I will share the results.

Considering we now have dozens of broadcasters who claim multipath improvement using SSB, I'd like to see Brian offer reasoning why it works, as compared to finding the opposite.

-Frank Foti

 

[TomZ]

This might not equate to that; but I used NewGrass Revival as a program source when I wanted pure tones and dynamics.

http://www.youtube.com/watch?v=qor_3az1YdU

 

[k6sti]

This is a prime reason why various simulations, done during algorithm development, do not play out in the real world. For years, I've done simulations on processed signals during the development stage. There have been times, where I got so deep into the development, using simulations, my thoughts truly believed the algorithm must work great because there was amazing peak control, and pure looking spectrum.

Sadly, many of those designs, fell flat on their face, when audio was applied, as we need to use it in the real world. Yet, the simulations looked incredible. I have also experienced the opposite, where a concept didn't look so good in simulation, but its performance was mind-blowing.

Tell you what, Frank. I promise not to hold the inadequacy of your simulations against you if you'll agree not to hold them against me.

The same holds true here with SSB, and multipath performance. Simulations are a good starting point, and barely half the story. I have already run many of Brian's tests, using a complete broadcast transmission chain, and the results are significantly different than what he keeps telling us. Upon completion of our testing, I will share the results.

Just what is it I keep telling you, Frank? I don't draw any conclusions in the writeup regarding multipath suppression. I just present simulation results.

Considering we now have dozens of broadcasters who claim multipath improvement using SSB, I'd like to see Brian offer reasoning why it works, as compared to finding the opposite.

And just where have I found the opposite? The simulation shows plenty of situations where SSB outperforms DSB. In one the SSB advantage is 79.8 dB. Isn't that enough?

Brian

 

[WNTIRadio]

I did read the report, a couple of times.

I think the flaw, at least from a practical "in the field" standpoint, which is where I make my living, is that your tests are too static. Is there a way to change the amplitude of the 5 reflections randomly? I wasn't clear before in that I know you did more than 1 reflection, what I meant to say was the constantly varying amplitude of those reflections. Your tests assume a fixed position, which is rarely the case in a car unless you're at a stoplight and decide to stay there. In fact, just as much as the amplitude varies of the reflections, so does it of the signal from the main lobe of the antenna to the receiver as the target is moving.

The question becomes how does SSB preform with varying amplitudes of both main signal and reflections, and reversals (for lack of a better word) when for a few moments the reflection becomes stronger than the main and back in a split second.

Like Frank was saying, some things in the lab environment stink and then work great in the real world. Some things look great in the lab and stink in the real world. As to the latter, 8VSB anyone? Especially 8VSB on a VHF channel.

I did enjoy the paper, however, I feel your tests are incomplete as they only present a static set of conditions which rarely exist "out there". The best simulation will attempt to simulate the real conditions and measure them... like a wind tunnel and mockup of a plane. You've put a lot of work into this but so far it's like one wing has been put in the tunnel, while the rest of the plane is still waiting to be attached to it.

 

[k6sti]

Is there a way to change the amplitude of the 5 reflections randomly?

Of course. But what insight would that provide?

The purpose of the simulation is analysis, not synthesis. I'm not trying to reproduce what a listener would hear in a particular situation for specific program material. I'm trying to figure out how multipath distortion behaves, what it is sensitive to, how it comes and goes, what it does with one stereo sideband rather than two. That's one reason I provide many whole-spectrum graphs at different delays and attenuations. If you scroll through the images, you can start to get a feeling for how multipath behaves, what makes it louder, what makes it go away, what changes its spectrum. Explore the later graphs to see how it varies with path delay and IF filtering. It's a highly nonlinear phenomenon that can't be characterized in a simple way. It turns simple tone modulation into highly variable, complex audio output. If I were to use multiple reflections, each with a random delay and amplitude, perhaps excited by actual program material instead of tones, how could you possibly unravel what's going on? I think tests like the one you propose are useful as a general check to make sure that results are reasonable. But they complicate things so much that any insight get buried.

If you'd like to pursue your idea, I'd be happy to provide the simulation source code. It runs quickly and is easy to modify. You can model any complex situation you can dream up.

Brian