Thursday, July 31, 2014

Spectrum Analyzer Configuration Experiments

Now that I have some compact filters with decent stop band rejection, I decided to try setting up a classic three stage IF conversion architecture.

The first stage is an R board providing attenuation of 0 - 30dB or +20dB of LNA gain.  The mixer stages are B boards with a mixer and synthesizer.  The first two filters are cascaded SAWs.  The last filter, ADC and IF amplifiers (not shown) are on an A board.   High side injection is used throughout, that is the first LO is set 1575MHz above the frequency of interest while the second stage is fixed at 1575+315MHz and the third stage at 315+10.75MHz.  The block diagram is shown below followed by a picture of the bench setup.
Three Stage Spectrum Analyzer Configuration

3 Stage Conversion Using Beagle Bone Black and BREC boards and filters.
The R board, first B, and A board are on a single Beagle Bone Black(BBB) at lower left.  The second and third down conversions are B boards on their own BBB in the center and right of image.  The test input source (a Prj114 USB PIC controlled ADF4351) can be seen between the first and second BBB.

A rough 3 stage calibration was conduct (so subsequent results are indeed in dBm) and as a quick wide band test, a 240MHz input was used with the first 1GHz scanned using Si.
The idea with 240MHz is that is is not a multiple of 25MHz (the PFD reference in any of the synthesizers).  The 25MHz spurious responses at markers 5,6, and 7 were not investigated at this time. Next a 7L12 calibration signal was used as the input.  This is a 50MHz signal with harmonics using a non compensated crystal.  The idea being that this is a non-ADF4351 based source and provides the lowest possible phase noise and spurs I can generate.  The wide band scan of that input is shown below.
The following is a 500kHz span about 50MHz of the same input signal.

Notice the spurs at the markers.  These were not expected.  A close up view of the input using a 50kHz span is captured in the next signal with envelope history on.
Basically, the filters and conversion are working as expected.  During calibration the anticipated gains were not as initially expected.  This is due to how I am using the LT5560 mixer.  I am using the device essentially single ended and over a wide range of frequencies in the first stage.  The input and output termination change over the frequency spectrum.  This is well documented in the part datasheet.  While it is a wide band part it was not intended to work simultaneously across its full range, rather, to be matched and balanced over a specific portion of its operating region.  This was a trade off I elected to make for power, cost and flexibility. The device has worked well and as stated - its just that I'm beginning to see the effects of those tradeoffs.

The spurs are a different story and are going to take some more effort to understand.  I always knew that the synthesizer would generate fractional spurs and could generate integer spurs and reference spurs.  I have been surprised by the quality of the spectra and have not taken the time to investigate the fine structure to this point.

Friday, July 18, 2014

Simple RF Source

As an aside while waiting for parts and boards I thought I would post a quick note about a previous project that worked well and turned out to be very useful as well as a good learning exercise with tools, instruments, and QFN construction techniques.

Prior to the availability of the Beagle Bone Black, my basic approach was to use 2"x2" boards with PIC microcontrollers and a USB interface.  In some of the posts there are references to a Prj114.  This was a simple ADF4351 with PIC control interface.  All together it costs about $30 (depending on how you count the PCB cost).  It uses the Analog Devices reference design as a starting point.  You can see the pads were not pulled out on the QFN part which caused me a great deal of trouble.  It is still useful as a quick and flexible standalone source.

Simple synthesizer using a PIC and ADF4351
Early on I started a practice of adding a small logo or drawing on the silk screen layer.  I read about some of the elaborate logos and illustrations people submitted with their OSHpark boards.  It turned out to be a quick visual identification when you have several purple PCBs lying around.  I continued it as it caught the attention of my family members and piqued their interest in what I was doing (even if they didn't care about phase noise measurements and PLL lock times).  The symbols are thought of at the last moment as I'm trying to get the PCB finalized.  Lacking any other inspiration and significant artistic ability, I generally choose something I can draw and involves a recent holiday, season, or activity.  In the case above Valentine's day.

Saturday, July 12, 2014

Cascaded SAW Filters (1575MHz)

The second SAW filter board was populated with 1575MHz filters. The EPCOS B3520 has a DCC6C footprint and can be used on the same PCB as the B3711  315MHz filter.  Evaluating the response turned out to be more difficult than anticipated due to leakage from the source into the measurement setup above 1GHz.  There was a lobing baseline present even with no input or no filter.  This behavior was present on both the 7L12, A-B boards, and SA0314.  To avoid it, rather than using a B board as a source, two B boards are mixed using 4Ghz and (4GHz - Fo) where Fo is the frequency being evaluated.  This way no synthesizer is at the frequency of interest.  The following are response scans from an A-B-R stack using Si compared to the datasheet.  The Si units are in absolute dBm while the datasheet is relative to source (dBc).  The limiting factor is the dynamic range of the A-B-R stack along with the limited output level of the source.
Narrow B3520 2X cascade response
Wide B3520 2X cascade response
The passband insertion loss was measured with the 7L12 and SA0314 and is as expected using 2X filter insertion and a 3dB coupling pad.  At roughly 1538 MHz there is a small spike in the response visible in the datasheet wideband response.  This appears to be -33dB below the passband.  In both the narrow and wide A-B-R scans this peak is only -40dB below the passband (in a 2X cascade).  If you eyeball a couple of points midway in the transition you do get a 2X value.  This tells me the rejection is not as much as expected particularly far outside of the passband.  All of the measurements were taken without shielding between stages. 

This has turned out to be more nuanced than the 315MHz measurement.  Given the test equipment at hand (sensitivity, source level control) very low signal levels are difficult to measure and without shields not valid.  The filter appears to work as expected even if I cannot measure every aspect of it I had hoped to.