Hi,

I am curious about the number of bits used error terms used for error correction and raw data stored inside PNA (what about the old 87xx?). Since we are considering moving from/to PNA, even old VNAs calibration data to an external PC, we would like transfer the data with exactly the number of bits that the instrument internally uses.

Thanks

lzw

I am curious about the number of bits used error terms used for error correction and raw data stored inside PNA (what about the old 87xx?). Since we are considering moving from/to PNA, even old VNAs calibration data to an external PC, we would like transfer the data with exactly the number of bits that the instrument internally uses.

Thanks

lzw

In the 8753/8720, is really a bit complicated. Back in the day, when memory was expensive and processing was really slow, the 8753 (and the 8510) used the trick that we called "shared exponent". Each of the the complex numbers was stored as a 16 bit real mantissa, 16 bit imaginary mantissa and a 16 bit exponent (if I remember correctly). Thus, the smallest angle that can be represented is arctan(1/32768). This limitation can actually be seen in time domain traces as little tiny immages down about 90 dB. In practice, the ADC used to digitize the signal is only 16 bits, so already the data processing is much more accurate than the data which was gathered. We called this "FORM1" data and we use it for moving data around inside the 8753, and I think someone once wrote a converter, but as it is 6 bytes instead of 8 bytes, it doesn't save much time.

This was in the day before co-processors, so having a common mantissa was a big time savings in complex math calculations.

For all practical problems, just use 32 bit. No reason to use 64 bit as the data is identical.