frequency shifter - BFO problems

[Haible Juergen]

Hi!

Over the weekend, I got the QBFO part of my frequncy shifter woking
(QBFO - quadrature beat frequency oscillator), but I ran into a very
strange and unpleasant phenomenon.

What I wanted is this, in short:

Fixed frequency oscillator (24kHz) has sine and cosine output
(simple phase shifter makes 90 degrees shift at fixed frequency).
90 degrees adjusted by scope in x-y mode, perfect circle.

VCO has only one (sine shaped) output, frequency range some 5kHz
... 50kHz.

Two 4-quadrant multipliers are used to mix the frequencies into  audio 
range, with the possibillity to go thru zero Hz. The two audio frequency
outputs have a 90degree shift now as well. (sin and cos). The upper
sidebands are cut by a filter)

sin_audio = fixed_sin * vco
cos_audio = fixed_cos * vco.

(frequency in both cases f_audio = f_vco - f_fixed)

Now the whole thing works very fine to a certain point: The outputs of
the
multipliers (very exactly adjusted, btw.) give also perfect circles on
the scope
in x-y mode. 

What I *expected* would be that when the difference frequeny between
the two oscillators becomes small, I see the beam rotating in one
direction,
then stop somewhere (on the circle ! ), when the difference is zero, and
start
rotating into the other direction.

What I *found* was a little different. The circles become elipses when
the
difference frequency is lower then some 20 or 30 Hz. Then you see it
slow
down, and change direction, but not on a circle anymore. In fact, when I
am
near zero, the beam wanders slowly across the the whole area inside
the circle, hoeizontal, vertical, in odd courves - anything. Just not
along the
line of the circle. 
This is very bad for the intended application. Ok, I would have expected
some changes in frequency, i.e. the beam slightly wandering around,
but always on the circle, so that I cet an almost constant amplitude
(sin**2 + cos**2).

Well, now I have found that I have made a bad design (and have to try a
better one), but what really puzzles me is *how* this effect is
produced.
The modulation works fine, if the two frequencies are are more than 30
Hz
apart. At the input of the multipliers, I have the desired 90 degree
phase shift,
in any case. But the output behaves strage at low frequencies.
I suspected the multipliers (RC4200 + resistor network for 4Q
operation),
so I checked with a pair of double balanced modulators (MC1496) - still
exactly the same thing.

I am really puzzled. Now could it be that a tiny phase jitter in the
24kHz
signals is transformed to a heavy phase jitter in the mixed down signal
??
(I haven't done the maths for this yet, but it could be fairly
reasonable
at least ...)
If so, then wouldn't I always get into trouble in the neighbourhood of
zero
Hz, regardless how much I improove the jitter of the HF-Oscillators?

Or is it something completely different?

I'd love to get some ideas, hints etc. what could be wrong.
I'd always love to get some experience shared from people who already
use frequency shifters (commerciall ones, or home built ones), if they
also have some unsteady level when the frequency shift is near zero.

Thanks for listening, many thanks in advance for any helpful hint!

JH.