Equalization and Symbol Rate

[cochenob]

Despite my number of posts on this board...I swear that I am happy with my NI devices.... 🙂

I noticed a peculiar thing today.  I am using some of the example programs to do some channel characterization.  The investigation is meant to be relatively simple.  I am using no baseband filtering, and since my transmitter and receiver are static I desire no equalization (in fact, right now, the up and down converters are connected via a short SMA as I do some "sanity checks".)

I am interested in utilizing as much of the 20MHz real time bandwidth as possible...so I went into the transmitter code (for example, the PSK Generate.VI located in the Modulation Toolkit Example folder) and hardwired 20MHz in for the bandwidth.  With no baseband filtering, the original code determines BW by scaling the symbols/sec by some number (10 I think) which would limit the symbols/sec to 1MSym/sec.

Anyway...with my 20Mhz BW...I set the symbol rate at 100kHz.  Using the PSK Demod VI in the same example directory, I observe the constellation and other measurements.  No problems here.  1Mhz symbol rate....OK.   Increase to 5Mhz...a little noisier, but OK.  Higher than that (say 10MHz symbol rate) and the MER and EVM start going way up.  Note that this is independent of the scheme (BPSK, QPSK, 8-PSK...whatever). 

I noticed that if I then turn on equalization (on both the Tx and Rx sides of course), that the MER and EVM goes way down....especially at the high symbol rates.

So my question....I thought equalization was only necessary if the channel response changes over time (multi-path,etc.).  Why then would equalization improve things when the symbol rate gets higher?  Is it possible I'm seeing the frequency/phase offset of the upconverter and downconverter LO's?   Or, maybe I'm just approaching the limit of the hardware and the equalization routines actually help clean up non-idealities of the hardware at those rates?  For the channel I'm characterizing, I am actually interested in seeing the effect of potential multipath, so I'd rather not use the equalization routines at this time.  Maybe I'm looking at having to make a tradeoff here?

---
Brandon

[Jesse_O]

Hello Brandon,

The behavior you are seeing is expected.  The equalization filter is used to correct the amplitude and phase of the signal so that after it passes though the 5610 you have a flat passband.  There is a section of the RFSG help called IF Response and Software Equalization and it can be found under Devices >> NI 5671 Overview >> Advanced Topics >> IF Response and Software Equalization.

I copied the typical magnitude response of the 5610 upconverter with the equalization filter disabled (upper graph) and with it enabled (lower graph) from the manual.  You can see it below.  I hope this answers your question.

Regards,

Jesse O.
Applications Engineering
National Instruments

Message Edited by Jesse O on 07-19-2007 02:59 PM

[cochenob]

Thanks. I tried looking for a document like this...but wasn't able to find it.  I'll look for the one you describe above.

Any specific info you could share as to why this occurs at the higher symbol rates?  (i.e. - what's the hardware limitation?)

[Jesse_O]

Hello,

For larger symbol rates, you span a larger frequency.  Having a larger frequency span without equalization allows for more distortion.  If you look at the picture I posted before, if you have a small bandwidth much equalization is not required, however over the entire 20MHz bandwidth much more equalization is performed to correct the signal.

To find the RF document I mentioned before, you can also find it here on page 26 and 27. 

Regards,

Jesse O.
Applications Engineering
National Instruments

[cochenob]

My appologies....I misinterpreted the graphs in my first reply.   This make complete sense now.

One last question....it seems completely plausible that I might use equalization on the transmitter end, but not the receiver end.  i.e. - adjust for non-idealities of the transmitter, and let the channel and receiver be what they may.  This would be in contrast to, say, the matched filter, where you must implement the same on both ends of the link.  Agreed?

[Jesse_O]

Hello,

The equalization performed on the transmitter is to correct the amplitude and phase of the output circuitry.  Once the data reaches the connector on the PXI module (5610) it is more flat and more correct than it would be without this equalization.  The receiver does not need any configuration to match this equalization like you would if you use a matched filter for both the transmitted and received data.  With matched filters you filter the transmit data to increase bandwidth efficiency and you will need to use this same filter on the receiver side to maximize SNR.   Where as with the equalization for the 5610 output path you are correcting the data to more closely match the ideal generated value.

Regards,

Jesse O.
Applications Engineering
National Instruments

[Andy_Hinde]

Using equalization on the Tx end will correct for non-idealities of the transmitter hardware - the PXI-567x in this case. You must use a matched filter that is the same as the pulse shaping filter like you say, but not equalizing on the Rx side will only have an effect for wide bandwidth signals and would only simulate impairments of the PXI-566x HW. I guess this would be plausible of your eventual receiver will have amplitude and phase variations on par with thePXI-566x.

Thanks,
Andy

[gitaa]

Hi,

when i generate a qpsk signal with carrier frequency 2.4 ghz and baseband signal is of 1MHZ and feed it to the ni 5660 psk demodulator, i can see the constellation plot correctly. but when with the same carrier frequency but baseband frequency of 2 mhz is used i'm not able to get correct constellation plot.

can u help me with this?

gitanjali

[USBarnes]

Gitanjali,

Thank you for posting. It seems that you have posted to a thread that quite old (has not been posted on since 2007). Likely you will get better assistance from the community if you post your question under a new thread. Feel free to add a link to this forum if you think the information is relevant.

That being said, I don't want to leave you without asking some more questions. What are you using to generate the qpsk signal? Could you also include some screenshots of what you see with the different settings you described and the code you are using for both generation (if you're doing generation in LabVIEW) and acquisition? Thank you for your time, Gitanjali.

[gitaa]

Sir ,

I'm using Xilinx FPGA board with analog devices FMC card(adfmcomms1 ebz) to generate qpsk signal .this signal i'm giving to my ni5660 as input and using ni5660 PSK demodulator to see the constellation plot to measure imbalance in I/Qsignal.

I used carrier wave as 2.4 GHz and baseband sine wave signal of 1MHZ then i'm getting proper constellation plot as shown in fig 1.

but when i change the baseband signal to 2MHZ i'm not able to get correct constellation plot as shown in Fig 2.