Test report on the adder module for VLBI phased array operation

M.Torres, P.Chavatte,  June, 2000
revised 03/2005

A  prototype module performing the addition of the signals of 6 antennas has been built. It makes extensive use of digital technology. Measurements on the reconstructed signal are presented. The dilution effect of the 10 kHz beacon signal (which is applied only on one antenna) is discussed.

Basic information on this technique and on the role of the module inside the system can be found on: http://www.iram.fr/IRAMFR/TA/backend/cor6A/#VLBI

A photograph of the prototype board can be seen on :  http://www.iram.fr/IRAMFR/TA/backend/gallery/Adder.jpg

1/ Principle

Each antenna delivers a 2-bit coded sample every 12.5 nanoseconds. Those samples are assigned the arithmetical values 0,1,2,3 which (coarsely) represent the analog voltage that has been sampled. Samples from all the antennas are added together, forming a number ranging from 0 to 18, depending on how many antennas are part of the game.
In order to compensate for this varying output, a digital potentiometer has been installed. The sum signal is later converted to analog, where a programmable analog lowpass filter reconstructs it with little degradation thanks to the oversampling effect.

The MarkIV formatter re-samples the reconstructed signal on 2-bits. The optimal level  for this sampling is specified to be 0 dBm but there is no built-in fine level control. The digital potentiometer can be used to optimize this re-sampling.  The resolution of the digital pot is such that the amplitude of the analog output  can be adjusted by 1% steps over a +3/-3 dB range.


2/ Block diagram

3/ Results

     1/ Bandwidth

The shapes of the reconstructed bands are given below for the values 4,8,and 16 MHz. At 16 MHz, a 3dB rolloff is observed which is explained partly by the sampling theory, and partly by hardware limitations in the D-toA block. Attenuators are automatically inserted  in order to keep total power constant when changing filters.

Amplitude response

2/ Dynamic range issues

The amplitude of the adder output depends on how many antennas are active, and also on the correlation of their signals.

We have measured the total power of the sum output versus the number of active antennas selected, with the digital potentiometer kept constant, and scaled it to one single active antenna.

a/ In the lab, where a single noise source is connected to all inputs, all 6 inputs are fully correlated, so it is their voltages that add, giving the practical rule of a 6 dB level raise when doubling the number of antennas.

Relative level variation for fully correlated inputs:

Active antenna number 1 2 3 4 5 6
Level (dB-measured) 0 5.96 9.51 12.0 13.9 15.5
Level (dB-predicted) 0 6.02 9.54 12.0 14.0 15.6

For this to happen, the phases have to be properly aligned.
In case two antennas are used and deliberately added with opposite phase, a null is produced. The depth of this null has been measured to -17.6 dB, which is an indication on how well two different channels are matched.

b/ At the observatory,  the antenna signals are essentially made out of receiver noise and can be considered fully independent. It is their powers that add, giving the the pratical rule of the output raising by 3dB when doubling the number of antennas. 

Relative level variation for uncorrelated inputs:

Active antenna number 1 2 3 4 5 6
Level (dB-measured) 0 2.97 4.75 6.03 7.00 7.81
Level (dB-predicted) 0 3.01 4.77 6.02 6.98 7.78

For example when switching from 6 antennas to one, the gain difference is 7.78dB, which means a voltage gain ratio of 2.45 (or SQRT(6)), which has to be applied in the digital pot to keep level constant.
For testing this, and since we don't have 6 physical noise sources, we have approximated them by applying huge delay differences on the 6 channels, using the built-in delay lines. Some additionnal decorrelation can be applied by setting the phase rotators each other a few kHz away.

3/ Beacon  10 kHz signal

The 1 MHz comb "phasecal" (see it here) signal is faintly coupled to the reference antenna. Its n-th line beats with the LO3 and the LO4 to deliver a XX.010 MHz  sinewave at the USB port of the relevant baseband converter. The following measurement on this port shows that very little phase noise is intoduced in the process. The view is taken arbitrarily on the 7.010 MHz line. In normal operation white noise from the receiver would set a floor at -75 dB. For this shot the noise source was shut off to enable visibility below this level.

Phase noise
10 kHz beacon spectral quality

When the noise source is switched on again, all elements of the analog channels are driven at their nominal power level. The application of the comb power on the reference channel is an increase of its operating power level. The shape of the comb signal is an extremely narrow pulse of high amplitude. This type of signal is far from the gaussian noise that the analog circuits are designed to handle. Second order distorsion takes place as shown:

IM2 in the analog chain seen on the BBC output

This distorsion (one line gives a family of lines, spaced by 10 kHz) is probably generated in the SSB mixers, where the subtle mechanism of unwanted sideband cancellation may be altered by the strong pulse of the comb. Anyway the level of the unwanted products is very low and will not impair the primary function of the comb which is to align the phases of the various BBC's.

It is interesting to see how faithfully this distorsion pattern (which has been now awarded the grade of signal ! ) is reconstructed after a couple of A-to-D and D-to-A conversions, as shows this shot taken on the output to formatter:

IM2 in the analog chain, seen on the analog sum output to formatter

The module includes a 10 kHz bandpass filter (~ 2 kHz wide) which can be used for extracting the pcal tone from the wideband noise of the reconstructed output and locally monitor its quality with an oscilloscope. The noise seen here is mainly amplitude noise.

10 kHz tone signal at monitor test point, ref antenna only.

The picture above is taken with the five other (non reference) antennas killed. Now if we add all six antennas, the total power being kept constant, the 10 kHz tone is diluted, and its S/N falls by 8 dB.  This is confirmed by the measurement below:

10 kHz tone with all antennas added

A spectrum analyser connected at the analog sum output displays exactly what the formatter will see:

Dilution of the 10 kHz signal delivered to the formatter:
Black trace: one antenna active
Green trace: all 6 antennas active

This display materializes the dilution effect, which is 10 log6 = 7.8 dB
One can note the lowfrequency rolloff, which is the overall response of the analog chain, which was designed at 5 kHz in order to preserve transmission of this line.