CC1120: Optimizing CC1120 modulation parameters for best reception, sensitivity, BER and range ...

Hi,

Ok thanks for the info. I assume (but correctly if I'm wrong) to calculate de modulation index H, you used the following general formula : H = (2 x dFpeak) / ( Sr x (M - 1))

where dFpeak is the outer deviation, Sr = symbol rate, and M = alphabet size (2 or 4 for 2FSK or 4FSK etc.)

If we take a 10K symbol rate, for a given fixed path attenuation, I measure BER minima at H = 0,666 for 4GFSK with a 30KHz filter. There also seems to be low BER errors at H = 1,0 approximately but with a larger filter set at 60KHz. Even though both seem to be "touching" on low levels of BER for same conditions, our preference in the real world would be the one with lower receiver filter bandwidth, around 30KHz, due to better noise performance and cross-talk between channels etc. Only slight drawback is possible errors due to the CC112X PLL crystal tolerance ppm, but there we chose a very low (2,5ppm) tolerance 40MHz TCXO, and the radios are usually calibrated one-by-one in production to reduce carrier variation to less than 1KHz ...

The eventual application is for En 300 220, and FCC 47 part 15.

As a sanity check, would these above values H = 0,666 and H = 1 make sense to you ? Would one over the other make more sense ? Should we even consider H > 1,50 ?

From your answer I assume the 4GFSK spectra would be very similar for the same modulation H = 1 or would the bandwidth be significantly more narrow ?

Also, I assume for 4GFSK on the CC112X, the inner deviation is 1/3 of the peak outer deviation by default, and probably fixed in hardware ??

Regards,

Mike Marin

Hi TA,

Further to your suggestion to use a modulation index H between 0,7 and 1,0, it would seem also necessary to match the receiver filter to the spectrum generated by the particular modulation index H. In other words, for a format of 4GFSK (to which my question relates in general) and a range of modulation indexes of 0,7 to 1,0 (which you advise) which would be more or less optimal in terms of minimum spectrum required for transmission, what is a matched filter range, (say like a “Rx filter bandwidth index” range) to go with the respective modulation indexes ?

Using too small a filter would lose demodulation capacity, and having too much filter bandwidth would degrade S/N as the filter is larger than necessary for demodulation, so there should be a compromise somewhere. Can you recommend this matching between modulation indexes H, and filter "Rx bandwidth indexes" for optimal reception (eventually in noisy environments noise would degrade the channel but …) for the CC112X radio transceivers ??

I think my filter BW question is the other half of the partially answered question, to getting optimal transmissions in eventually noisy environments. We would appreciate your advice on this matter for this particular device please.

Regards,

M Marin

TER said:

For RX required RX BW:
----
Required RX filter bandwidth can be approximated as :Signal BW + 4*ppm xtal*Frequency of operation , where Signal BW of = Data rate + 2 x frequency deviation.

Example: 38.4 kbps data rate and +/-19.2 kHz deviation gives Signal BW of38.4 + 2 x 19.2 = 76.8 kHz

For modulation index m= 1, the frequency separation, which is 2 x frequency deviation, is equal to the data rate. I.e m = 2 x frequency deviation / data rate
---
Here the signal bandwidth is an approximation. To find the actual RX BW:
- Use Matlab or similar to calculate it
- Measure it

Hi TER,

Thanks for your suggestions. I think for 1-FSK this is a good approach and would certainly use it when necessary, however for 4-GFSK these "of the back of an envelope" calculations don't give us enough accuracy for what we're looking at. Take for instance the graph above for 4FSK, with a modulation index H =  1 and 50Ks/s the deviation dF = 75KHz, then gives 50 + 2 x 75 = 200 KHz is the calculated signal bandwidth using the carson's rule estimate.

Then for a 2,5ppm TCXO crystal at 869 MHz say,  I calculate a Rx BW = 200 + 4 * 2.17 = 208.7 KHz allowing for crystal tolerances and aging. The MATLAB spectrum for 4-FSK show a 99% signal bandwidth of 174,805 KHz and at 90% of 150,000 KHz. I would assume with certainty the spectrum is even more efficient for Gaussian 4-FSK (but do not know by how much), so all I am saying is that going by simple carson's rule for 4GFSK can easily lead us to over-estimating the required filter bandwidth.

I thought it may be a good start to ask Ti directly regarding the choice of Rx BW they would recommend for the CC112X chip, other than what is available in most literature. So if the estimation is to coarse I guess measuring it could be another good way to go....

Regards, Mike

Data rate + 2 x frequency deviation works fairly good on 2-FSK doing quick estimates but for other modulations simulations or measurements are the way to go to find the signal bandwidth. 

TER said:

Not sure if I understand your last post. What I have indicated you can simulate is the modulation bandwidth since the 2-GFSK and 4-GFSK shape from CC112x is fairly close to text book. The RX filter can be regarded as brick wall since the order of the filter is high so no need to simulate the RX filter.

Hi TER,

Yes thanks, I think it helps knowing to treat the filter as a "brick wall filter".

thanks and regards,

Mike

Mike, 

I just re-read my post from earlier and I wrote 2FSK, where I actually meant to write 2GFSK. I am sorry about that, so I will now post all 4 combinations for your reference. Also note that my initial post was using a BT=0.5 which is not supported by these device, so for this repost I have modified the Gaussian BT=1.0 as per what the device actually transmit.

1. 100kbps 2FSK
   
2. 100kbps 2GFSK
   
3. 100kbps (50 k symbols/s) 4FSK
   
4. 100 kbps (50ksymbols/sec) 4GFSK
   

Sorry about the initial error, hope this helps.

Regards,
/TA

Hi TA,

Yes thanks for clearing that up for me. There is a significant difference between the 4FSK and 4GFSK for the 99% power bandwidth, almost 10% less bandwidth. Good to know this as we are trying to minimize the occupied bandwidth without degrading the channel.

I thought I'd do some typical BER measurements with a 544 bits pseudo randome cyclic pattern on our custome transceiver at a 18kS/s and here are some typical results for various filters. Not immediately obvious the relation beween optimal deviation dF, Rx Filter BW and least bit-errors.

I guess wie'll be doing this for all relevant speeds and filters. Unfortunately we don't  have a good Matlab model of the Ti transceivers, and if we were to build one we'd have no way to verify it is actually representative, so you're right, measurements are best way afterall to find these "sweet-spots" if we want to ensure optimal configuration.

Cheers, M Marin