Does anybody here has any experience with FRA?

[saman]

I'm designing a FRA(Frequency Response Analyzer). Maximum frequency that
I need is about 1MHz. Now for multiplication of response wave to
reference Sin(a) and Cos(a) I have two ideas, either use a very fast
MDAC which I couldn't find a suitable one till now or use an analog
multiplier.

Your helps are welcomed

 

[saman]

I'm designing a FRA(Frequency Response Analyzer). Maximum frequency that
I need is about 1MHz. Now for multiplication of response wave to
reference Sin(a) and Cos(a) I have two ideas, either use a very fast
MDAC which I couldn't find a suitable one till now or use an analog
multiplier.

Your helps are welcomed

If I were you I'd use a pair of double-balanced diode-ring mixers e.g.
Mini-Circuits SBL-1 or one of the newer siblings. Generate sin/cosine
signals (e.g. I and Q), amplify, feed through a 50-ohm attenuator pad
(say, 6 dB or so), and into the "LO" ports of two SBL-1 mixers.
Similarly, amplify/buffer the incoming signal, split it and pad it,
and feed to the "RF" ports of the two mixers.

Feed the "IF" port of each mixer to a 50-ohm buffer (e.g.
grounded-base amp) and feed it through a diplexor/filter which has a
bandpass as wide or narrow as you wish, then measure the resulting
output.

[The padding and buffering is required to ensure that each of the
three ports on the mixer "sees" a 50-ohm impedance to any signals
coming out of the port - this is necessary for proper operation.]

If I were doing this, I'd consider using an IF of 3.579545 MHz, as
this would allow the use of an inexpensive filter using cheap NTSC
colorburst crystals (I think there probably some resonant transformers
pre-wound for this frequency, too), and it's high enough above your
upper frequency limit that a simple low-pass filter prior to mixing
would eliminate any image problem.

I imagine that you can do this with just one mixer and one LO
signal... I don't think you need the full quadrature pair in order to
simply measure the signal amplitude.

The device I'm designing is supposed to work from near DC to up to 1MHz.
DDS is used for very accurate frequency generation.

 

[saman]

How near to DC? Single-digit Hz, fractions thereof, tens of Hz,
hundreds of Hz?

These diode-ring mixers are frequently used in direct-conversion
receiver applications, where the signal coming out of the "IF" port
consists of audio frequency information.

The frequency responses of the three ports of an SBL-1 (e.g.) vary...
e.g. the LO and RF ports are spec'ed to allow 1 MHz through 600 MHz,
while the IF port is spec'ed as "DC to 500 MHz".

Your application would probably be best treated as a direct-conversion
modulator / up-converter. Feed your test signal into the IF port
(which has good low-frequency response), feed your DDS signal to the
LO port, and take the output from the RF port and filter it and do a
level-detection on it.

Another type of mixer/multiplier you could consider is the "H-mode"
mixer, or related FET-based switching mixers. One interesting type
uses a family of commonly-available analog bus switch ICs (it's even
been done with 4066 CMOS parts).

You can design a mixer/multipler with this sort of switch which
doesn't require transformer coupling... you can use a simple
capacitive coupling / bias network to bias the incoming signal up into
the center of the IC's voltage range. The lower frequency limit would
then be set by the size of the coupling capacitor, and the impedance
of the bias network and the load.

lowest frequency is 1mHz(1 milli Hz).
Let me explain a bit. In FRA frequency of response wave and reference
wave is the same the only thing that differ is their phase.
The mixer should not have any AC coupling.
Also phase accuracy is critical here and especially two reference waves
should have exactly 90 degree phase shift.

After multiplication result will be integrated.
Any DC component can introduce a huge error here.
One of other challenges that I have is how to remove DC component in
such low frequencies. One solution that I thought about is to use a bias
DAC for that purpose.

 

[saman]

See if you can find a copy of "Experimental Methods in RF Design" to
borrow, it has a good treatment of mixers in direct conversion and
other applications.

Then take a look at SDR- Software Defined Radio. ie a Tayloe mixer.

Then get a look at schematics for commercial LOCK-IN Amplifiers which
often have very accurate mixing techniques be definition.

I personally like the idea of a phase detector or doing it with a
microprocessor, but I have no idea of the spex you wish to meet.

I have done a similar project with Lock-In amplifiers. I've used ADG202
analog switch there. Yes Lock-In Amplifiers are superb considering their
S/N ratio. Lock-Ins are good for as low as 10Hz but I need even lower
frequencies.

 

[saman]

Diode mixers are good for much higher frequencies than you need;
probably a simple Gilbert cell is suitable for your purposes. It
takes three
matched pairs of transistors, and the output is high impedance,
suitable for lots of IF filter inputs.

That kind of multiplier also supplies gain, if you were going to need
that...

The operational circuits called 'analog multiplier' are overkill for a
task like this. Switching multipliers (diode mixers and H-bridge
switches) will require you to deal with odd harmonics somehow.

Gilbert Cell sounds interesting.
One of characteristics that is very important in final accuracy is THD.

Almost all of you have suggested analog multiplication. How about
digital multiplication using MDAC ?

I have played for awhile with mathlab and tried to determine what
parameters can determine final accuracy.
I tried to multiply two waves with the same frequency that had phase
shift and integrate the result. I found THD can cause errors in
measurement.

The reason that I also inclined toward analog multiplication is just that.

 

[saman]

whit3rd wrote:

The reason that I also inclined toward analog multiplication is just that.

I just want to add this that though if output from DDS passes a low pass
filter before going to an analog multiplication it improves THD
significantly but because as I said it is absolutely critical that
sin(a) and cos(a) reference waves have exactly 90 degree phase shift.
so if I pass sin(a) and cos(a) from low pass filters after they came out
of DDS there is a possibility that phase shift of the two waves may not
be exactly the same because errors in components value. while this is
really negligible because sampling frequency of DDS is about 100
MSPS(100 time more than maximum frequency that I need) and 3db knee is
far from frequency of the wave but still in worst case it may introduce
a few degrees of phase shift.
If I don't use low pass filter after DDS for analog multiplication, then
THD of both analog multiplication and digital multiplication are the same.

 

[saman]

I was reading on the web last night about FRA's (by which you can
surmise I am a complete novice.) Could you hit your circuit with
white noise and look at the response? If you just look at the
response you'll get only amplitude information. But, I think, if you
also record the input noise waveform you can tease out the pahse
information as well.

George H.

That is not noise really, it may looks like noise but it is a very
precisely generated waveform that consists of multiple frequencies.
after applying this waveform the response is read and response of each
frequency is extracted using FFT.

 

[saman]

Multiplying D/A converter and a sine/cosine lookup table does sound
like a good way to do the mixing; two issues arise, though. First,
the D/A will lookup (maybe 1000) values, and at 1 MHz that means
you need a new lookup every nanosecond. At 1 Hz it means
you only update the value every millisecond. It's hard to scale this
solution to a wide range of target frequencies.

The second issue, is that there is a glitch when the D/A converter
updates, and you don't want to integrate the glitches; your
integrators,
then, have to be programmable to be 'dead' for a short period,
or you have to characterize the switching transients. It can be
made to work, of course, like by using extra converters so
each 'dead' time is covered by an alternate unit that was updated
earlier.

glitch of D/A converter is a very good point that may influence final
accuracy. thanks for pointing it out.