A Board IF Amplifier Gains

The Beagle Bone Black A-board 1MSPS ADC gain stages were characterized.  Measurements were taken using a 3mVpp input sine wave. The IF gain can be configured by the OpAmp used and feedback resistors employed.  The following table provides two measured configurations.  The gain of a stage (S0/S1) is controlled by the GPIO (G0/G1) selecting the mux setting controlling the feedback resistor.  The first stage input impedance is chosen to match the filter termination required (300 ohms) while the second stage is small (47 ohms) to allow a smaller feedback resistor (and noise).  In the table, the first stage feedback values of 3k or 4.7k are used in one configuration and 300 or 3k in the other.  Second stage values of 470 or 1k are used in both configurations.  For lower first stage gain configurations a unity gain stable OpAmp is used.  The table also shows the deviation from the simple gain calculation incorporating filter attenuation.

Unit	G0	G1	Calculated S0 gain (dB)	Calculated S1 gain (dB)	Calculated Module gain (db)	Measured gain (dB)	Measured-Calculated (dB)
1	0	0	20	20	37	37	-0.1
1	0	1	20	26	43	44	0.9
1	1	0	23	20	40	41	1.0
1	1	1	23	26	46	48	2.2
2	0	0	0	20	17	18	0.8
2	0	1	0	26	23	24	0.7
2	1	0	20	20	37	29	-8.4
2	1	1	20	26	43	35	-7.5
Calculated module gain is S0+S1-3dB for filter
Unit 1 uses AD4895 for both S0 and S1
Unit 1 settings are: S0=(3k || 4.7k)/300, S1=(470 || 1k)/47
Unit 2 uses AD4897 for S0 and ADA4895 for S1
Unit 2 settings are: S0=(300 || 3k)/300, S1=(470 || 1k)/47
Measurements using oscilloscope with 3mVpp input at 10.7MHz

 
The values match expectations (particularly given the error margin involved using a scope with small signals). The compensated OpAmp deviating more from the simple gain calculation is anticipated.

A-B-C VHF Narrowband FM

Corrected some problems with synthesizer PLL charge pump and fractional N settings.  Incorrect configuration led to a low phase frequency detector pulse rate which caused a spreading of the higher frequency LO.  VCO voltage would droop then shoot back with over compensation - the net result looking like a low frequency FM modulation on the LO.  With a single stage conversion the IF was ~120MHz which goes through a 32 way division masking the problem.  With a two stage conversion the higher frequency LO clearly showed the line broadening (FM wide and narrow band stations sounded like you were in a tunnel - discernible but sort of muffled ).

Below is a screen shot of NOAA/NWS VHF local weather broadcast.  The software package shows the entire 500kHz span available on the right.  Two software VFOs are used to look at seperate narrow band FM channels within the span - those are on the left.  The upper left is the NOAA NBFM, lower left is a channel tuned off to a quiet portion of the spectrum.  Audio quality is the same as previous captures for this station using A-B SDR configuration.

Antenna configuration is the same as previous work - 19" of 30AWG just hanging, no ground plane. Both B & C board amplifiers enabled, A board is unit #2 (lower noise configuration).

Updated A and C Board Response

Addressed an error in the sampled rms estimate software used in the response self scanning.  The updated results are below.  The C board is closer to expected.  The high floor still needs investigation.  The A board response has a tilt to it and is less flat than expected.  To me this seems to indicate either a termination mismatch and/or poor ground on one side of the filter.  Unfortunately I need to take the board stack apart to evaluate and address - which if I had used the gold plated connectors would not be a big deal.  The large headers on the A board are gold and are amazingly smooth.  The pass through headers on the middle B board are tin plated.  With the insertion/extraction force involved I get a little nervous with the two layer boards and the synthesizer QFN and where it sits on the board (fracturing working solder joints).  I'll defer this until I get some more measurement mileage out of the existing configuration.  Pop for the gold plated pins - its worth it.

While on the topic of mistakes, yes the C board filter response is too wide to provide effective image rejection at the A board IF directly (I'll admit it - I misread the data sheet, wasn't closely looking at the horizontal scale.  Some of the other parts considered for this stage had sufficiently narrow bandwidth to work at the A IF, however, they were in the 800-900 MHz range. I got greedy and didn't watch the details).   Will need to work through the non-instrumentation application options from another IF stage to external cavity filters.

C Board First Tests

The C board is finally constructed and ready for evaluation.  The first BC board (unit #1 used in previous tests in a B configuration) was re-worked as a C board, the second BC board (unit #2 recently populated) configured as a B board.  In addition, both boards were mounted on A board unit #2.  This A board is configured with lower gain (and noise) than the A board used in previous tests.

After a cursory check of the synthesizer, mixer and amplifier I decided to make sure the SAW filter is acting as expected.   Since the mixers are single ended input and output the first LO will appear on the output of the first mixer.  This is used to estimate the filter response and act as a built in test.  The figure below summarizes the measurement setup. LO0 is swept across the frequency range of interest with LO1 set at 10.75MHz above LO0.  The C board IF output is connected to a 7L12 spectrum analyzer for manual measurements and absolute level evaluation or an A board for multistep scans. ( yes this would all be better done with automated test equipment, none of which I have or can afford ;-)

 Below is the filter response from the data sheet.

 

The following are two excel graphs from the Beagle Bone Black application software (it steps the frequencies, grabs the power estimate at the step and puts out a CSV file which is readily graphed in Excel).  The first matches the data sheet frequency span and center, while the second doubles the frequency span and catches the outer skirts.

In addition to the above, similar measurements were taken manually with the 7L12 for comparison and validation (albeit with many fewer points).  Those results match the above.

In comparing these results to the data sheet a couple of issues are apparent.  First the response on the upper side of the pass band is attenuated and not flat.  This matches well with the 7L12 measurements using different final IFs.  I believe this is an interaction between the amplifier, filter, and mixers.  The mixers are configured as single ended input and outputs with resistive termination (i.e. resistors for current source/sink).  This was done on purpose for simplicity and broad band application.  The down side is that the impedance presented at the inputs and outputs at higher frequencies is not constant or purely resistive may be skewing the filter response.

The second artifact present is that the stop band attenuation level is less than specified by the data sheet.  Again, these results match with manual 7L12 measurements.  I believe there is a certain amount of bleed through of LO0 into the second stage mixer around the filter.  This provides a lower power level floor at a given step which the filter cannot attenuate below.  Further measurements are required (this would not be a surprise as the RF isolation between the boards is not what it would be in a shielded environment).

Board Stackup

The design leverages a stacking approach.  There were several factors which led to this including:
a) Desire to stay within a small form factor (board costs, workable two layer boards)
b) Desire to avoid expensive board-board connectors (cost and ease of hand construction)
c) Desire to minimize cables (cost, hand construction, tediousness and lack of tooling)
The following stack is used with readily available parts.