Re: Multiplex decoder nearly rewired.



John Byrns wrote:

> In article <437C5E56.59006049@xxxxxxxxxxxxxxxxxx>, Patrick Turner
> <info@xxxxxxxxxxxxxxxxxx> wrote:
>
> > I saw the filter in the Quad circuit and i don't know what F it is for, 55khz,
> > or 19kHz, since there is no 19kHz filter indicated.
>
> It is a low pass filter, presumably cutting off starting somewhere around
> 55 kHz, with a null tuned to 67 kHz according to the alignment
> instructions in the QUAD manual. The 67 kHz null was necessary in the US
> to eliminate the 67 kHz SCA sub carrier components, plus it gives the
> filter a sharper cutoff, although at the expense of ultimate attenuation
> at higher frequencies. The QUAD stereo mpx design does not include a 19
> kHz filter, 19 kHz filters were uncommon in FM stereo tuners prior to the
> introduction of Dolby noise reduction into cassette recorders.

We don't have 67kHz subcarriers in Oz, so no need to filter out what isn't there.
But its looks like a bandstop filter, not a low pass filter.
The parallel LC circuit is in series with the composite signal going to the matrix.

Probably you are right about no 19kHz filters.

Often they are placed at the output of each channel.
But early tuners with onboard mpx units with 19kHz notch filters on each channel
were rare.
There is an unmeasurable amount of 19kHz in the outputs of my MPX unit.


>
> > > > Good that you have mentioned the 55kHz filter.
> > > > When I looked at the Quad decoder I didn't spot this filter.
> > > > I will try such a filter to see if it reduces the much larger hiss
> level in
> > > > stereo
> > > > that I am getting compared to mono, even when no modulation is applied
> > > > at my transmitter.
> > >
> > > You might want to look into combining your 19 kHz pilot filter with the
> > > low pass filter as was done in the Marantz 10B. A former Marantz employee
> > > said that with proper design the phase shifts introduced by the 19 kHz
> > > pilot filter can be used to compensate for the phase shifts introduced by
> > > the low pass filter into the upper sidebands of the 38 kHz DSBSC sub
> > > carrier.
> >
> > But filtering to separate the L+R from the DSB is not required.
> > Scott also don't separate the composite.
>
> Yes, both the Scott circuits and the QUAD circuit keep the composite
> signal together and do not separate the L+R from the DSB, however it is
> still necessary in the US to eliminate the 67 kHz SCA sub carrier, as well
> as high frequency random noise resulting from the FM demodulation
> process. Today audiophiles feel that 19 kHz filters are a necessity least
> the residual 19 kHz bother their golden ears. The 19 kHz filter is
> commonly inserted in the two audio paths coming out of the decoder, but
> some times it is placed before the decoder as you have done. In the
> design of the famed 10B Marantz combined the 19 kHz null with the low pass
> function and the 67 kHz SCA null into a single integrated filter design,
> which has been claimed to offer improved phase compensation.
>
> > The whole idea of the Quad circuit and some othr similar ones, and including
> > the Scott circuits is to keep all the components of the composite signal
> together,
> > so that good separation is possible across a wide band.
> > The basic idea is to simply add the missing 38khz carrier to the
> existing composite
> > from the
> > ratio detector, or whatever detector is used.
>
> Yes that is how it is usually, but not always, done. The low pass and 67
> kHz null filters used between the ratio detector and the mpx decoder cause
> phase shifts or time delays that are not constant across the band up to 53
> kHz occupied by the stereo composite signal. This nonlinear time delay
> causes a loss of stereo separation, keep in mind that modern audiophiles
> are not content with the 20 to 30 dB of separation that you are willing to
> settle for, and demand separation figures in the 60 dB range least the
> "sound stage" be compromised.
>
> Marantz simply observed that by combining the 19 kHz null with the 67 kHz
> null, and the 54 kHz low pass in a single coordinated package, they were
> able to make the phase shifts due to the 19 kHz null filter compensate for
> the phase shifts introduced by the 67 kHz null and low pass filter, or so
> a former Marantz employee has claimed. I haven't gone through the math to
> verify this, but I assume that the idea is that the phase shift of the 19
> kHz null at the upper end of the L+R audio band is designed to complement
> the phase effects on the demodulated L-R audio resulting from the
> combination of the phase shift of the 67 kHz null and LPF on the upper
> sideband of the 38 kHz sub carrier and the 19 kHz null on the lower
> sideband of the 38 kHz sub carrier. The idea being that the L+R and L-R
> audio would line up correctly in the decoder ring and subsequent auxiliary
> matrix. This would not however imply that the time delay would be
> constant vs. frequency for the demodulated left and right audio signals.
> This is just my speculation as to what Marantz was trying to do with their
> filter, I have not gone through the analysis, so my speculation could
> easily be all or partly wrong. Comments about the design of the Marantz
> composite filter from anyone that knows more would be welcomed.

I am getting very good separation which is over 23dB at 1kHz, and which improoves
up to 16kHz.
It can actually be increased to 35dB if one fiddles with it and there is a double
null situation;
I go go for the middle between the nulls to get the best stabilisation of the 38kHz
oscillator; tuning the 38kHz oscillator secondary away from 38kHz provokes
synchronization to stop, and a beat not is heard off lock.
25dB sep at 5kHz is OK.
And the sep tends to drift a little with temp of the coils, and adjustment is fine
with only
1/8 of a turn of the tranny slug able to reduce the sep from 25dB to 10 dB.

So once I am done I will wax the cores.


>
> > The phase of the carrier relative to the subcarrier waves is easily
> > swayed to line up exactly.
>
> Yes, you can sway the phase of the carrier relative to the sub carrier
> waves so that it effectively matches the phase of the original suppressed
> carrier, but I don't believe that necessarily implies that the phase of
> the demodulated L-R audio and the L+R audio will correctly line up in
> phase to give maximum separation, especially across the entire audio band,
> that depends on the constancy of the time delays in the composite filter
> comprised of the 19 kHz null filter, 53 kHz low pass filter, and the 67
> kHz null filter.

It works well in my case.

The type of notch filter I use for 19kHz has a very narrow notch,
and very small phase effects on the signals.
Had I used a simpler parallel LC circuit, either a low value R feeding a series type

LC, or high value R feeding a parallel LC filter to get a high Q, there would be
other serious phase effect in the two bands of interest, the AF and the 38kHz band.

The bridged T notch filter with L&C is by far the *only* sort of filter that should
be used
where a single F is to be eliminated.
broader filters to stop a whole band centred on 67 kHz can be the simpler
LC type as Quad used.
It won't have much effect below 53 kHz.


>
> > Any attempt to filter out the DSB and thus gain the L-R signal so that
> it may be
> > applied
> > to LPF filtered phases of L+R, and -L-R results in poor separation at HF
> due to
> > inevitable phase shifts caused by what must be more than first orer filters.
> > I tried all this in many efforts but always the phase shifts prevented
> more than
> > 12dB sep at 10kHz, and
> > almost no sep at 13kHz.
> >
> > The phase effects of the 19kHz notch filter between 23kHz and 53 kHz and
> between
> > 20Hz and 15kHz is utterly negligible.
>
> I'm not sure what you mean by "utterly negligible", clearly something is
> causing your poor high frequency separation even if it isn't your 19 kHz
> filter.

You can shunt out the 19kHz notch filter with little effect on the recovered L&R
except that there is 19kHz present in the signals.

Its a large size component if not filtered because the pilot is about
10% of the maximum audio modulation.

> In any case the phase effects of the 67 kHz null and 53 kHz low
> pass filters are not negligible, I take it what Marantz was trying to do
> in the 10B is to also make the phase effects of the 19 kHz null filter non
> negligible and then take advantage of the phase effects of the 19 kHz null
> to compensate for the effects in the decoded audio resulting from the 67
> kHz null and LPF.

I have not seen the Marantz schematic.
Nor have I worked on one. And trying to theorize about
the exact phase response of such things is rather difficult.

>
>
> > It can be shunted out and all that does is in my similar to Quad circuit
> is dump
> > a whole lot of 19khz in each L&R output, plus probable IMD products.
> >
> > Much of what we listen to, especially the HF portions, is at a level below the
> > pilot tone levels.
> > IN my case I am able to banaish the damn pilot tone after a cathode
> follower, and
> > thus
> > drive drive the diodes with a low impedance.
> > There *is* a small effect of the 19kHz filter elements after the follower
> > but that is easily compensated for with a cap on the cathode R of the
> gain before
> > the CF.
> > We are not looking for much compensation, about 3dB max.
>
> I wouldn't call 3 dB an "utterly negligible" effect! Are you sure it is
> the "small effect of the 19kHz filter" that your "compensation" capacitor
> is correcting for? That 3 dB figure is suspiciously close to the
> theoretical 3.9 dB difference in level between the L+R and L-R signals in
> the FM stereo composite signal.

The passage of the composite signal from ratio detector always
gives some reduction of DSB levels.
Perhaps the bandpass shape of the ratio detector tranny
reduce the DSB a little.
Anyway, the addition of a cap across the cathode R of the composite
SET gain stage works well to slightly boost the DSB and tweak its phase.



>
>
> > > > > 4.) It isn't clear how you are compensating for the fact that the fact
> > > > > that the L-R sub carrier signal is effectively transmitted at a lower
> > > > > level than is the main channel L+R signal? QUAD used an auxiliary
> matrix
> > > > > to perform this function, feeding inverted L+R audio into the outputs of
> > > > > the decoder to cancel some of the L+R signal.
> > > >
> > > > Indeed you are correct about the Quad additional matrix.
> > > >
> > > > But I saw that as a compensation network because separation levels
> of L and R
> > > > modulation
> > > > recovered tend to be poor because the amplifying of the ratio
> detector signals
> > > >
> > > > tends to lessen separation; the composite signal envelope is mauled by the
> > > > RC couplings and gain variations between the audio frequencies and the
> > > > subcarrier F.
> > > >
> > > >
> > > > > There are at least two
> > > > > other was to perform this function, are you following the QUAD
> lead, or is
> > > > > the "compensation" capacitor in the cathode circuit of the input
> amplifier
> > > > > a crude solution to this problem.?
> > > >
> > > > The compensation cap isn't as crude as you think; it tweaks the
> phase of the
> > > > supressed carrier double sideband signal and slightly increase the gain at
> > > > around 38kHz.
> > >
> > > It is a crude approach to equalizing the differing levels of the L+R and
> > > 38 kHz L-R signals.
> >
> > Well my system works very well to give over 30db of separation, and is simple.
>
> I thought you said the separation was only 20 dB at higher frequencies?

I measured it last night again, and got 25dB at 10kHz.

Its never less than 20dB until you get to about 20Hz when sep has fallen to
about 18dB which doesn't matter since bass F are in-phase anyway.


>
>
> > > > Then you should find the separtaion from the diode ring are either L or R,
> > > > without any need for Quad's method of compensation.
> >
> > The diode ring seems to cause some loss of separation and there would only be
> > perfect
> > sep if there was perfect detection, as theory suggests it should happen,
> but it just
> > doesn't occur.
> > The CF and the oscillator have to provide the power to the matrix, and
> > the some losses seem to occur.
>
> There is a theoretical "problem" with the FM stereo signal that you
> haven't considered yet that is responsible for most of this loss of
> separation, even if the diode ring were perfect you would still see the
> effect.

So what is the "theoretical problem" i have not considered?

>
>
> > > > This is understood if you just modulate the FM signal on R audio
> channel only
> > > > with nothing on the L channel.
> > > > The magnitude of the L+R signal = that of the modulation carried by
> the DSB
> > > > signal,
> > > > which is L-R.
> > > > The signal applied to one side of the diode ring will then appear to be
> > > > an AM 38khz carrier but which has the top of the envelope modulated
> with the R
> > > > channel
> > > > signal, and the bottom should be the L channel modulation, so it
> should be a
> > > > flat line of mod
> > > > because there is no L channel mod in tis test case.
> > > >
> > > > The other side of the diode ring has an inverted version of the
> other side,
> > > > flat line on top, R mod on the bottom.
> > > >
> > > > The levels of L-R transmitted at the station must be fixed to be the
> same for
> > > > all stations for all tuners to give the same separation.
> > >
> > > The point you are missing is that the peak amplitudes of the L+R audio and
> > > the 38 kHz sub carrier are the same, but the demodulated L-R audio is
> > > lower in level than the L+R audio after it is multiplied by the 38 kHz
> > > switching function. The math to show this is relatively simple.
> >
> > Let me see if I have it straight.
> > If you modulate just one channel with a 1 kHz sine wave, 1Vrms, then the
> the L+R
> > audio wave from the ratio detector
> > has an amplitude equal to the amplitude of the L-R modulation
> > contained by the double sideband signal, ie, the envelope modulation
> shape has an
> > amplitude of 1Vrms.
> >
> > The composite signal when viewed will then have an amplitude of 2Vrms,
> consisting
> > of the +ve 1/2 1kHz waves filled with 38kHz waves, between its arch and
> 0V, and
> > then followed by
> > the -ve 1/2 waves with 38kHz filling between the 0V line and the arch of
> the sine
> > wave.
>
> Yes, if I followed your explanation correctly the signal is as you have
> described it. To correctly demodulate this signal without an auxiliary
> matrix or "compensation" capacitor to boost the sub carrier gain, it is
> necessary to use an impulse sampling function, where as with your high
> amplitude sine wave you are effectively using a square wave sampling
> function, which results in the L+R component of the signal having a higher
> amplitude than the demodulated L-R component, hence poor separation.

I am not sure what you mean by impulse samplig function.
But yes, with a 56 p-p V applied across the diode + R ring,
the adjusments to the cap charge levels cannot be instantaneous.
If you look at the Quad circuit there are 4 x 22k plus diodes in a "square" pattern,
then there
are a pair of 12k taken from the diodes to a cap of 4,000pF, C11, and C12 on the
Quad schematic.
The 4,000pF seems to be an incorrect value.
I set up my diodes + R with R = 4 x 27k and the "build out" R = 10k, and C = 900pF.
which I found just right to give low ripple F or switching noise *and*
the de-emphasis of 50uS.
The outputs from the caps are either L or R, and in my case go
to CF buffers; I found loading the 900pFs with any low R following worsened the
separation.
So there must be an interaction between the caps being charged to make the audio
signal and the
circuit driving it.



>
>
> > There is a strong tendency for this signal to try to loose the amplitude
> > relationships between DSB quantity and L+R audio quantity.
>
> It is more than just some vague "strong tendency", the lower amplitude of
> the recovered DSB quantity is a direct result of the way the FM stereo
> signal is mathematically defined.
>
> > So if the 38kHz wave is accurately applied to the composite,
> > the resulting mainly 38khz wave form will have the L signal on the top
> > of it, and R signal on the bottom of it.
>
> Yes, one channel will appear on the upper peaks of the 38 kHz carrier,
> while the other will appear on the lower peaks, but that doesn't imply
> that you can completely separate the two channels with a square wave
> switching function of the type you are using,

But there are no square waves used anywhere.

> for that you need an impulse
> type sampler. When you multiply the composite signal by a 38 kHz
> switching function as you are, the L+R signal is attenuated by a factor of
> 1/2, while the L-R signal demodulated from the 38 kHz side bands is
> attenuated by a factor of 1/PI. This is a difference in level of 3.9 dB
> and is suspiciously close to the approximately 3 dB compensation you say
> your "compensation" capacitor introduces.

I am not sure how you figure all that.
Anyway, my system works well, so i plan to stay with it.

>
>
> > Simple diode detection could be used to retrieve the signal but the use of the
> > matrix
> > and balancing results in a staircase stepped wave at the caps off the
> diodes, not a
> > saw tooth
> > "ripple voltage" like one sees in a power supply, or diode & RC detector
> > in an AM set.
> > Filtering the steps out is easier than filtering out 38kHz of ripple.
>
> It sounds to me like "ripple" is just another word for "steps"?

There is a vast difference.

In a detector for AM radio at 455kHz, the ripple voltage at a typical 100pF cap
is a saw tooth wave; its ok because its all easily filtered out since
455kHz is well above audio.
But in the diode matrix used in a decoder, the two R and two diodes each 1/2 of the
matrix
both conduct depending which way the flow of current is in the 38kHz.
The cap with AF information is either being charged quickly on +ve going audio peaks
or
discharged quickly on -ve going audio peaks. There is a very slow time constant
used for the discharge of the charge at the caps since there isn't any need for the
caps
to discharge at all between the fast charges applied by the 38kHz waves.
So a stepped wave is the result.

Its very like the D to A process where a stepped wave id first produced which must
be converted to
a sine wave by removing the steps by filtering. Except a CD player does it 44.1kHz,
and the FM decoder does it at 38kHz.
Its would have been cleaner to have the process at say 76kHz,
and the pilot at 38kHz but that
would have meant less multiplexing ability and some more critical
matrixing circuitry.

Its possible to have the creation of the audio via a simple diode & CR detector
but the saw tooth wave resulting would have a higher amplitude than the
step wave does, and a little more phase shift in the detection process.
Either detector process does create a phase lag in the recovered audio of around
90 degrees at 15kHz.

That is a bother with nearly all radio audio signals.
Lotsa extra phase lag occurs in the HF signals.

But its nothing compared to what is done in nearly every studio to music
which has been much manipulated and equalised.
There is almost no recorded phase coherent music to be listened to.
Our ears don't mind.


>
>
> > > 2.) The BA1404 signal generator may be generating a nonstandard stereo
> > > signal that helps to compensate for separation problems in your stereo
> > > decoder.
> >
> > Nope, after years of testing other tuners, the BA1404 provides a signal
> to them all
> > which
> > could be separated very well.
>
> I am sure you are correct here, and that the BA1404 includes a
> compensation circuit although the data *** doesn't mention it. I think
> I was confused about the conclusion I had drawn about the BA1404 many
> years ago, thinking about it again I think the problem was that the BA1404
> puts out a square wave 38 kHz DSB signal, which if feed through a very
> wide band tuner, or directly into a mpx decoder without a 53 kHz low pass
> filter, will impact the separation because there will be L-R sidebands
> around the third harmonic of 38 kHz which if not filtered out at some
> point will be demodulated by a square wave decoder increasing the level of
> the demodulated L-R signal with a consequent impact on the separation.
> This may partly explain why you only required 3 dB of compensation rather
> than theoretical 3.9 dB, or even more when potential high frequency losses
> in a real circuit are considered.

Since inserting a filter into the BA1404 transmitter circuit to try
to convert the 19kHz square wave to a sine wave, no change to separation occured.

But the DSB waves at the ratio detector show up as sine waves, not square waves.


>
>
> > > What you are describing is the design used in Scott's early decoders,
> > > check this link for a later Scott tube decoder design:
> > >
> > > http://hhscott.com/pdf/345.pdf
> >
> > My god, 4.6Mb.
> > Can't you reduce the file size of stuff like this?
>
> It wasn't me, I just provided the link.

OK.

>
>
> > Its also a little hard to follow, but it seems little different in principles
> > to what have been used commonly except the 38Hz seems to be
> > created by amplifiying the rectified 19kHz, rather than have a 38kHz which is
> > then syncronised. Either will work fine, especially in good signal areas.
> >
> > The locked oscillator is better imho, because
> > any amplitude changes in the 19khz, or audio spuriae
> > at 19kHz will not alter the amplitude of the 38khz so badly.
> > Its very important for fidelity that the 38khz has a steady amplitude.
>
> I think if you look again you will see that there is positive feedback
> from the plate circuit to the grid, making it an oscillator, unless what
> you are saying is that there isn't enough positive feedback to make it an
> oscillator, and the positive feedback is only there to increase the gain?

OK, maybe there is, as I said its hard to follow, but it would be possible to
make the 19kHz amp which also doubles the F within it to make 38kHz
to also act as an oscillator.

>
>
> Notice also that in the later Scott designs like this one the out of phase
> "compensation" signal for the auxiliary matrix is taken from the "bottom"
> side of a floating ratio detector.
>
> > The front end of my set is ex-trio with 2 x 6AQ8, 2 x 6BA6(IF), 6AU6 limiter,
> > and the rest are in the MPX.
> >
> > It could have another limiter; some FM sets did have two limiters,
>
> Doesn't your set use a ratio detector? The 6AU6 plus the ratio detector
> make a total of two limiters.

The ratio detector does tend to reject AM, if that is what you mean.

I did some measurements last night.

The hiss noise from the decoder is dependant on limiting in the
IF part of the set.

Strong stations generate -13V of AVC voltage applied to the 3 IF tubes including the
limiter.

My sig gene only makes enough RF to generate -4V AVC, and the SNR
of hiss to signal level of 4.2Vrms output which is about 1/2 the
maximum daytime audio signal levels received on many stations
is about -50dB.
But when the sig gene is directly connected to the RF input the AVC generated
is about -12V, and the SNR is -66 dB.

I tried comparing it to the Audio Reflex tuner and found that it was better
compared to mine when mine was receiving low power signals, but worse
with high power signals, so as long as the station signal strength
is high my set will have a better SNR than the Audio Reflex.
Hum levels were the same for both tuners, and not much above the hiss levels,
and I think due to hum generated at the BA1404.

I reduced the gain of the composite signal gain amp and
to give 1/2 the gain, so 2.1Vrms is produced instaed of 4.2Vrms of audio at the
outputs
and found the SNR only got worse by slightly more than the gain reduction so
I will go back to having a 6DJ8 with a gain of 5 instead of 2.5.

I will place a resistance divider between the last LPF filter on the L&R
outputs and the final CF output buffers to reduce the signal levels to
about 1Vrms and match the filtered output from the AM section of the set.

The method I am using is producing much more output voltage than the system
I used first in the decoder at my website, which may R.I.P, since it is a
way of doing things which has serious flaws.

Patrick Turner.





>
>
> Regards,
>
> John Byrns
>
> Surf my web pages at, http://users.rcn.com/jbyrns/

.