6 Gsps ADC Interleave Test - CASPER

6 Gsps ADC Interleave Test
Rick Raffanti
Sept 10, 2007
The purpose of this test was to interleave two ADC083000 boards and measure the ADC board
frequency response. Each ADC board sampled at 3 Gsps, for an effective sample rate of 6
Gsps. The two ADC boards (Assembly R271, revA), were plugged into a Virtex 5 board (”Roach
1.0”, revA), with FPGA type XC5VSX50TFF1136 installed.
I drove the two ADC clock inputs with 1.5GHz through a ZAPDQ-2-S quadrature splitter (so one
is delayed ¼ cycle at 1.5GHz, thus ½ cycle at 3GHz). I drove the two ADC inputs through a
ZFRSC-42-S+ resistive splitter. I carefully matched cable lengths, and used all SMA
connections. Clock frequency of the ADC interface is 375MHz, for a 750MHz DDR data rate.
Test setup shown in photo below.

First I acquired 128 sample records and plotted them at various input frequencies:
250
200
150
100
66MHz
166MHz
266MHz
50
0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126
250
200
150
100
566MHz
866MHz
1166MHz
50
0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126
200
180
160
140
120
100
80
2066MHz
2566MHz
3066MHz
60
40
20
0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96 101 106 111 116 121 126

Next I acquired 1024 sample records and computed rms amplitude vs frequency. I used two
different RF generators for the input, an HP8616A (only goes 1.8GHz and up) and a PTS3200
(1MHz to 3.2GHz). I calibrated the PTS3200 output using an HP432A RF power meter. The red
plot below shows the calibrated data.
Sine Sweep, Mini Roach 6GSPS
2
0
-2
dB rel 66MHz
-4
-6
-8
-10
-12
-14
PTS3200 generator
HP8616A generator
PTS3200, calibrated
1 10 freq, 100MHz
1000 10000
Conclusions:
From the 266MHz plot it’s clear that the differential sampling phase is off by something like 50ps.
This will distort the waveforms, especially at higher frequencies, but shouldn’t affect the rms
values.
Analog bandwidth falls off rapidly above 2.5GHz, probably due to the ADC083000 response itself
(National warned us that the 3GHz specified in the data sheet was not actually achieved).
Between 1GHz and 2.5GHz response is bumpy, maybe due to balun or PCB issues.