Here are a selection of audio spectrum displays of various radios and software. I have tried to get the scales to match each other but there are a few exceptions that I have indicated in the red overlay on each image.
First an RTL dongle running at 2048k sample rate. Not much obvious in the way of de-emphasis, maybe it is using the incorrect broadcast standards of 50/75/100uS.
With HDSDR you have to change the audio output rate to correctly change the receive bandwidth. The actual bandwidth adjustment slider makes very little difference.
With SDR Console, I'm not sure what is going on, but I couldn't seem to get the results I wanted and N-FM produced all sorts of odd traces.
So it would seem with SDR's a lot depends upon the software in use, the sample rates and various other configuration items.
Let's have a look at some more traditional radios. The first is an Icom-7000 transceiver, which is 'sort of' analogue, but it does use some DSP. It is very difficult to spot the subtle differences between the different RF bandwidths.
Finally some old analogue 25KHz channel spaced handheld radio's.
Bottom line, out of all of them I trust the Analogue radio's de-emphasis curves the most, and these do seem to be quite severe. By eye they seem to be about 15dB down at 3KHz. Which is in basic agreement with Johns predicted curve using Octave showing -17.5dB at 3KHz.
The SDR software is all over the place, with different bandwidths to the settings, various slope shapes and unwanted artifacts.
Assuming that various radios also have different implementations of pre-emphasis, and may also have some other audio processing to shape and compress the speech in order to make it more punchy. Then I think Waynes observations of the off-air audio may be expanded by all of the above.
I think the best bet is to stick to the theoretically defined NBFM de-emphasis curve and apply that. Then test to see if the actual result is thr same as predicted. Which is pretty much what we are doing.
10m was fairly open this morning here in North Carolina and I managed to capture short audio recordings from both the Airspy and Kiwi on the same frequency. The recordings were from the SDR#'s audio recorder and Kiwi's recorder respectively. Both bandwidths were set to 12 KHz, and the LF+ filter was selected on the Kiwi.
To me, the SDR#'s audio is much more intelligible than the Kiwi, but the low frequency corner is too high and the "lows" are gone on the SDR#. The Kiwi has better low frequency response, but the highs are gone and it sounds "muffled". Here's the frequency response of the two receivers again, then the audio recordings:
I can't figure out in SDR# how to adjust the filter bandwidth properly.
Unless you have the filter audio box ticked in the audio control panel, you get full bandwidth audio regardless of the filter bandwidth setting in the radio control panel.
If you tick the box you can vary the upper frequency limit using the filter bandwidth setting in the radio control panel, but there seems to be a fixed amount of LF filtering that you can't adjust.
Either way SDR# seem to have the wrong de-emphasis curve for NFBFM, as it is nowhere near -15dB at 3KHz, so I think they are using the incorrect WBFM values of 50/75 or 100uS.
The filter tick box doesn't seem to do anything in the WBFM mode, and the IF spectrum display in SDR# also seems to be incorrect in this mode too.
Basically I just don't trust SDR# to have the correct values.
I would go across to the Airspy IO group and ask about this, but I have crossed paths with Youssef before, and I really can't be bothered to inflict that sort of pain on myself again.
Jim Lill WA2ZKD suggested trying SDRconsole with the Airspy, so I got the latest version (3.2) and gave it a shot. The curve looks more like the Kiwi, and sounds similar to the Kiwi, but there's still a bit more high frequency response with a modulated voice FM signal on 10m. Unfortunately, I wasn't quick enough to record them on both receivers and right now there are no signals to use. Here's the no-signal spectrum of SDRconsole 3.2 with the Airspy Discovery HF+ as the hardware:
If I hear more activity and can grab a recording I'll post it.
Here's a no-signal FM audio spectra comparison from build 1.577 (presented above) to build 1578. There were no signals on 10m late in the evening to listen to.
I have done a bit more research and the problem seems to be that there isn't really a specification for Narrow Band FM pre/de-emphasis, other than the curve should have a 6dB / Octave or 20dB/Decade slope.
I have looked through various ETSI, ITU and FCC specifications and they all concentrate on the transmitter performance and limiting the level of radiated emission in the adjacent channel.
Some transmitters intended for commercial applications, where there is a defined channel spacing, will sometimes have an additional low pass 'anti splatter' filter, typically at about 3kHz, placed after tpre-emphasis is applied. This is to preven the gradually rising level of modulation from creeping into the adjacent channel. However a lot of radios don't incorporate this additional filter, and they do transmit outside of the desired channel bandwidth.
There is a great deal of variation between manufacturers about the frequency at which the 6dB per Octave pre-emphasis should start. Some start at very low frequencies and just continue at 6dB per ocatve from 100Hz all the way up, without really having a properly defined corner frequency. Others can start with the per-emphasis at frequencies as high as 1kHz. This variation makes a big difference to the amount of boost applied at the higher frequencies.
The transmitter audio band may also have a low pass filter at around 200Hz so tha any speech doesn't interfere with CTCSS or DCS signalling tones, and this can also affect the transmitted frequency range.
The same issues apply to the de-emphasis applied in the receiver, and this too varies from manufacturer to manufacturer. Ideally if you used the same radios from one manufacturer, they would have matched transmit and receive characteristics, and this would probably be true in most commercial organisations. However it can be problematic where a mix of radios is used, such as in VHF marine radio, or amateur radio, where different radios are required to communicate with each other.
With the KiWi there are some other factors at play.
If I measure the demodulated audio with a broadband level meter, and listening to the noise on a clear frequency.
De-emphasis off on on +lf
NBFM 0dB -8dB -6dB
NNFM -2dB -8dB -7dB
The overall level of broadband white noise is reduced as we would expect. However subjectively the audio level drops considerably, as our hearing doesn't have a flat characteristic amplitude frequency curve.
In addition the audio levels produced by the KiWi when different types of demodulation are used also varies.
For example if we were to receive an AM signal with 400Hz tone used as modulation, and used a modulation depth of 30% (a standard used for testing). We would obtain a certain output level.
With amplitude modulated signal, the level of recovered audio will depend upon the RF gain, but for most of the time AGC will be used to try and keep the recovered audio within certain limits, as signals fade up and down in strength. So for these tests I have used a moderate signal level of around -60dBm, which should ensure that the KiWi is always operating within the AGC window. However the AGC will tend to compensate if different AM modulation depths are used.
Using the 30% modulated AM signal, demodulated as AM, as a 0dB reference for the recovered audio.
AM 400Hz tone 30% modulation AF = 0dB
SAM 400Hz tone 30% modulation AF = 0dB
USB (one sideband) 400Hz tone 30% modulation AF = +11dB
CW (one sideband) 400Hz tone 30% modulation AF = +10dB
NBFM 400Hz tone 5.0KHz FM deviation at 30% modulation +/- 1.5 kHz AF no de-emph= -5dB
NNFM 400Hz tone 2.5KHz FM deviation at 30% modulation +/- 0.75 kHz AF no de-emph= -10dB
Interestingly SAM produces a high level audio output when a NBFM signal is applied, which I was not expecting.
SAM 400Hz tone 5.0KHz FM deviation at 30% modulation +/- 1.5 kHz AF no de-emph= +13dB
SAM 400Hz tone 2.5KHz FM deviation at 30% modulation +/- 0.75 kHz AF no de-emph= +10dB
It's apparent that the different modes produce different audio output levels, some of which would be expected. However there is a >20dB range between USB and NNFM, which explains why I have to move the the volume control slider all the way up to to 100% when listening to NNFM, especially when the perceived level is further reduced by switching the de-emphasis on.
Could the differing audio levels be compensated for in some way ?
Maybe a +6dB boost needs to be applied when de-emphasis is switched on, and a 6dB reduction when SSB or CW is in use.
I think the issue with what Wayne is hearing, is down to the frequency at which the de-emphasis kicks in.
John has implemented a true definition of 6dB per octave with the break point occurring at 400Hz.
With that break point, and de-emphasis on, the level at 1KHz measures at around -6.5dB, and at 2kHz it is -12.6dB.
If the break point was moved to 1KHz, then the attenuation at 2KHz would be about 6dB less.
Maybe a compromise is to set the break point somewhere in between 400Hz and 1KHz, perhaps even 500 or 600Hz would make a difference, so that the 6dB per ocatve rule is still true, but the audio frequency roll-off is not quite so severe, and can better accomodate transmitters that have a dubious pre-emphasis characteristic ?
Incidentally, I also checked the aliasing problem I previously mentioned. There is still some aliasing present, but with the de-emphasis switched on, it is not as problematic as it was before.
This has been a very interesting, but somewhat frustrating investigation, and it's clear that a lot of radios and SDR's don't behave as they should with respect to the filtering, demodulation and recovered audio.
But now, thanks to Johns efforts, the KiWi seem to be one of the better ones.
I've now had a chance to listen to a mix of different NBFM off-air signals.
To me the de-emphasised audio sound slightly muffled, but not excessively so. The UK CB band that uses NBFM radios specifically designed for the UK market, sound acceptable, but other CB's, typically multi mode radios being used by French or Italian stations vary considerably.
I also listened to some amateur NBFM transmissions on the 2m band. These were typically using +/-2.5kHz deviation, but the operators were not speaking that loudly, so the overall level of modulation was quiet low. As a result they did sound far too quiet, especially in comparison to the other modes. In fact on one station, I had to turn all the PC volume settings up to maximum, in order to hear what was being said. Maybe John could include a volume setting of 11 (as per Nigel Tufnel, the guitarist in Spinal Tap) to help cope with these instances ?
One of the things about NBFM with a very low modulation index is that you can also listen to then using SSB.
This is because the NBFM signal almost is the same as AM, except the phase of the modulation sidebands relative to the carrier is different. When using SSB, the carrier is not required as a reference for the signal to be demodulated, so it's possible to demodulate the modulation as if it was an AM signal.
When switching between NNFM and USB, the level of the recovered audio increases by about 10dB, which is about the same as I observed when I was using a signal generator as a source.
This has convinced me even more that the difference in the level of recovered audio when swittching between modes, and also de-emph on / off, is contributing to the perceived difficulty in listening to NNFM with de-emph on.
Comments
This has opened up a whole can of worms.
Here are a selection of audio spectrum displays of various radios and software. I have tried to get the scales to match each other but there are a few exceptions that I have indicated in the red overlay on each image.
First an RTL dongle running at 2048k sample rate. Not much obvious in the way of de-emphasis, maybe it is using the incorrect broadcast standards of 50/75/100uS.
With HDSDR you have to change the audio output rate to correctly change the receive bandwidth. The actual bandwidth adjustment slider makes very little difference.
With SDR Console, I'm not sure what is going on, but I couldn't seem to get the results I wanted and N-FM produced all sorts of odd traces.
So it would seem with SDR's a lot depends upon the software in use, the sample rates and various other configuration items.
Let's have a look at some more traditional radios. The first is an Icom-7000 transceiver, which is 'sort of' analogue, but it does use some DSP. It is very difficult to spot the subtle differences between the different RF bandwidths.
Finally some old analogue 25KHz channel spaced handheld radio's.
Bottom line, out of all of them I trust the Analogue radio's de-emphasis curves the most, and these do seem to be quite severe. By eye they seem to be about 15dB down at 3KHz. Which is in basic agreement with Johns predicted curve using Octave showing -17.5dB at 3KHz.
The SDR software is all over the place, with different bandwidths to the settings, various slope shapes and unwanted artifacts.
Assuming that various radios also have different implementations of pre-emphasis, and may also have some other audio processing to shape and compress the speech in order to make it more punchy. Then I think Waynes observations of the off-air audio may be expanded by all of the above.
I think the best bet is to stick to the theoretically defined NBFM de-emphasis curve and apply that. Then test to see if the actual result is thr same as predicted. Which is pretty much what we are doing.
Regards,
Martin
10m was fairly open this morning here in North Carolina and I managed to capture short audio recordings from both the Airspy and Kiwi on the same frequency. The recordings were from the SDR#'s audio recorder and Kiwi's recorder respectively. Both bandwidths were set to 12 KHz, and the LF+ filter was selected on the Kiwi.
To me, the SDR#'s audio is much more intelligible than the Kiwi, but the low frequency corner is too high and the "lows" are gone on the SDR#. The Kiwi has better low frequency response, but the highs are gone and it sounds "muffled". Here's the frequency response of the two receivers again, then the audio recordings:
Airspy Audio (zipped)
Kiwi Audio (zipped)
Hi Wayne,
I can't figure out in SDR# how to adjust the filter bandwidth properly.
Unless you have the filter audio box ticked in the audio control panel, you get full bandwidth audio regardless of the filter bandwidth setting in the radio control panel.
If you tick the box you can vary the upper frequency limit using the filter bandwidth setting in the radio control panel, but there seems to be a fixed amount of LF filtering that you can't adjust.
Either way SDR# seem to have the wrong de-emphasis curve for NFBFM, as it is nowhere near -15dB at 3KHz, so I think they are using the incorrect WBFM values of 50/75 or 100uS.
The filter tick box doesn't seem to do anything in the WBFM mode, and the IF spectrum display in SDR# also seems to be incorrect in this mode too.
Basically I just don't trust SDR# to have the correct values.
I would go across to the Airspy IO group and ask about this, but I have crossed paths with Youssef before, and I really can't be bothered to inflict that sort of pain on myself again.
Regards,
Martin
Jim Lill WA2ZKD suggested trying SDRconsole with the Airspy, so I got the latest version (3.2) and gave it a shot. The curve looks more like the Kiwi, and sounds similar to the Kiwi, but there's still a bit more high frequency response with a modulated voice FM signal on 10m. Unfortunately, I wasn't quick enough to record them on both receivers and right now there are no signals to use. Here's the no-signal spectrum of SDRconsole 3.2 with the Airspy Discovery HF+ as the hardware:
If I hear more activity and can grab a recording I'll post it.
Hi Wayne,
If you use SDR console, can you open out the spectrum analyser to up to 20KHz.
I'd like to know if you see the NFM spectrum with a second band of stuff between 12 & 18kHz that t can be seen in one of my previous plots.
Regards,
Martin
Looks like the cutoff wasn't long enough.
Here's a no-signal FM audio spectra comparison from build 1.577 (presented above) to build 1578. There were no signals on 10m late in the evening to listen to.
Hi Wayne,
OK that's good and is what I'm seeing too.
I have done a bit more research and the problem seems to be that there isn't really a specification for Narrow Band FM pre/de-emphasis, other than the curve should have a 6dB / Octave or 20dB/Decade slope.
I have looked through various ETSI, ITU and FCC specifications and they all concentrate on the transmitter performance and limiting the level of radiated emission in the adjacent channel.
Some transmitters intended for commercial applications, where there is a defined channel spacing, will sometimes have an additional low pass 'anti splatter' filter, typically at about 3kHz, placed after tpre-emphasis is applied. This is to preven the gradually rising level of modulation from creeping into the adjacent channel. However a lot of radios don't incorporate this additional filter, and they do transmit outside of the desired channel bandwidth.
There is a great deal of variation between manufacturers about the frequency at which the 6dB per Octave pre-emphasis should start. Some start at very low frequencies and just continue at 6dB per ocatve from 100Hz all the way up, without really having a properly defined corner frequency. Others can start with the per-emphasis at frequencies as high as 1kHz. This variation makes a big difference to the amount of boost applied at the higher frequencies.
The transmitter audio band may also have a low pass filter at around 200Hz so tha any speech doesn't interfere with CTCSS or DCS signalling tones, and this can also affect the transmitted frequency range.
The same issues apply to the de-emphasis applied in the receiver, and this too varies from manufacturer to manufacturer. Ideally if you used the same radios from one manufacturer, they would have matched transmit and receive characteristics, and this would probably be true in most commercial organisations. However it can be problematic where a mix of radios is used, such as in VHF marine radio, or amateur radio, where different radios are required to communicate with each other.
With the KiWi there are some other factors at play.
If I measure the demodulated audio with a broadband level meter, and listening to the noise on a clear frequency.
De-emphasis off on on +lf
NBFM 0dB -8dB -6dB
NNFM -2dB -8dB -7dB
The overall level of broadband white noise is reduced as we would expect. However subjectively the audio level drops considerably, as our hearing doesn't have a flat characteristic amplitude frequency curve.
In addition the audio levels produced by the KiWi when different types of demodulation are used also varies.
For example if we were to receive an AM signal with 400Hz tone used as modulation, and used a modulation depth of 30% (a standard used for testing). We would obtain a certain output level.
With amplitude modulated signal, the level of recovered audio will depend upon the RF gain, but for most of the time AGC will be used to try and keep the recovered audio within certain limits, as signals fade up and down in strength. So for these tests I have used a moderate signal level of around -60dBm, which should ensure that the KiWi is always operating within the AGC window. However the AGC will tend to compensate if different AM modulation depths are used.
Using the 30% modulated AM signal, demodulated as AM, as a 0dB reference for the recovered audio.
AM 400Hz tone 30% modulation AF = 0dB
SAM 400Hz tone 30% modulation AF = 0dB
USB (one sideband) 400Hz tone 30% modulation AF = +11dB
CW (one sideband) 400Hz tone 30% modulation AF = +10dB
NBFM 400Hz tone 5.0KHz FM deviation at 30% modulation +/- 1.5 kHz AF no de-emph= -5dB
NNFM 400Hz tone 2.5KHz FM deviation at 30% modulation +/- 0.75 kHz AF no de-emph= -10dB
Interestingly SAM produces a high level audio output when a NBFM signal is applied, which I was not expecting.
SAM 400Hz tone 5.0KHz FM deviation at 30% modulation +/- 1.5 kHz AF no de-emph= +13dB
SAM 400Hz tone 2.5KHz FM deviation at 30% modulation +/- 0.75 kHz AF no de-emph= +10dB
It's apparent that the different modes produce different audio output levels, some of which would be expected. However there is a >20dB range between USB and NNFM, which explains why I have to move the the volume control slider all the way up to to 100% when listening to NNFM, especially when the perceived level is further reduced by switching the de-emphasis on.
Could the differing audio levels be compensated for in some way ?
Maybe a +6dB boost needs to be applied when de-emphasis is switched on, and a 6dB reduction when SSB or CW is in use.
Regards,
Martin
I think the issue with what Wayne is hearing, is down to the frequency at which the de-emphasis kicks in.
John has implemented a true definition of 6dB per octave with the break point occurring at 400Hz.
With that break point, and de-emphasis on, the level at 1KHz measures at around -6.5dB, and at 2kHz it is -12.6dB.
If the break point was moved to 1KHz, then the attenuation at 2KHz would be about 6dB less.
Maybe a compromise is to set the break point somewhere in between 400Hz and 1KHz, perhaps even 500 or 600Hz would make a difference, so that the 6dB per ocatve rule is still true, but the audio frequency roll-off is not quite so severe, and can better accomodate transmitters that have a dubious pre-emphasis characteristic ?
Incidentally, I also checked the aliasing problem I previously mentioned. There is still some aliasing present, but with the de-emphasis switched on, it is not as problematic as it was before.
This has been a very interesting, but somewhat frustrating investigation, and it's clear that a lot of radios and SDR's don't behave as they should with respect to the filtering, demodulation and recovered audio.
But now, thanks to Johns efforts, the KiWi seem to be one of the better ones.
Regards,
Martin
I've now had a chance to listen to a mix of different NBFM off-air signals.
To me the de-emphasised audio sound slightly muffled, but not excessively so. The UK CB band that uses NBFM radios specifically designed for the UK market, sound acceptable, but other CB's, typically multi mode radios being used by French or Italian stations vary considerably.
I also listened to some amateur NBFM transmissions on the 2m band. These were typically using +/-2.5kHz deviation, but the operators were not speaking that loudly, so the overall level of modulation was quiet low. As a result they did sound far too quiet, especially in comparison to the other modes. In fact on one station, I had to turn all the PC volume settings up to maximum, in order to hear what was being said. Maybe John could include a volume setting of 11 (as per Nigel Tufnel, the guitarist in Spinal Tap) to help cope with these instances ?
One of the things about NBFM with a very low modulation index is that you can also listen to then using SSB.
This is because the NBFM signal almost is the same as AM, except the phase of the modulation sidebands relative to the carrier is different. When using SSB, the carrier is not required as a reference for the signal to be demodulated, so it's possible to demodulate the modulation as if it was an AM signal.
When switching between NNFM and USB, the level of the recovered audio increases by about 10dB, which is about the same as I observed when I was using a signal generator as a source.
This has convinced me even more that the difference in the level of recovered audio when swittching between modes, and also de-emph on / off, is contributing to the perceived difficulty in listening to NNFM with de-emph on.
Regards,
Martin