Excessive frequency deviation for DMR handsets?

A Hytera DMR handset that I am working with has a frequency deviation of 2300-2700 Hz. This is higher than the expected 1944.0 Hz frequency deviation specified in the DMR specification (ETSI TS 102 361-1). Is it expected that low cost handsets have excessive frequency deviation? I'm wondering how the receiver performs synchronization when the frequency deviation can be so far out of range?

• I/Q samples are collected with an SDR
• This question is not about carrier frequency offset but rather the FSK frequency deviation
• To determine the deviation, I used an FM demodulator then measured the distance between minimum and maximum deviation.

Update 1:

• the frequency deviation is closer to 2100-2400 Hz after further measurement.
• how do you derive the frequency deviation from your recording, exactly? Commented Aug 26, 2023 at 9:55
• FM demod oversampled data, then measure the max to min outer deviation. Commented Aug 26, 2023 at 12:43

Is it expected that low cost handsets have excessive frequency deviation?

No. Such a transmitter wouldn't work with standards-compliant receivers. Most symbol decisions would end up being wrong, assuming sensible decisions regions.

The "zero hypothesis" here is that your transmitter works compliant to the standard, and your measurement methodology is off. (Sad, but true!)

To determine the deviation, I used an FM demodulator then measured the distance between minimum and maximum deviation.

That's sadly not a proper way to do that!

Remember that in actual FM transmissions, the audio message is integrated before it because the frequency deviation; so an FM demod will give you the derivative of the instantaneous frequency. Now, the derivative of the instantaneous frequency (for the purely theoretical "perfect" FM demodulator) is given by the symbol rate together with the differences in frequency, not just by the frequencies.

Even if you have a demodulator that leaves out the derivative (that in the end just being a bit of a highpass to the audio), the relationship between instantaneous frequency and output of the demodulator is a proportional one, but you have a degree of freedom in designing your demodulator's frequency sensitivity. So, you need to check that very closely what you're doing here.

• The software based FM demod I am using takes into account the sample rate: fm = np.angle(x[1:] * np.conj(x[:-1]))*fs/ (2*np.pi), where x is the oversampled complex baseband FSK signal and fs is the sample rate (integer multiple of symbol rate). This does give the proper frequency deviation. I have used this technique for many FSK signals. Commented Aug 26, 2023 at 14:41
• that's not an FM demod, but a frequency demodulator (I know, this is very nitpicky, sorry, but when I hear "FM", that's referring to a specific method of modulation for analog signals); but, hm! do you do that during the say center 10% of the symbol period, or over the full signal? I ask because of the overshoot you might get through pulse shaping prior to FSK on the transmit side. Commented Aug 26, 2023 at 14:52
• The estimation if performed over the full signal Commented Aug 26, 2023 at 15:01
• huh! So, sanity checks are in order: do we have an easy way to check whether the symbol rate is as it should be? (if in doubt, approximate the discrete derivative of your fm, look for the zero crossings, that's where the derivative of the signal is zero, and that should be at the symbol instants) Commented Aug 26, 2023 at 15:14
• That's a clever way to resolve symbol timing! How does that work if there are multiple symbols of the same value in a row? You won't have a 0 crossing. Commented Aug 26, 2023 at 15:25

Verification through simulation of a DMR burst shows that the excessive frequency deviation is caused by the overshoot from the pulse shape filter ($$\alpha$$=0.20). The DMR modulator uses a RRC at the transmitter and receiver which can cause the deviation to go beyond 1944 Hz.

Here are the ideal simulated frequency deviations after the RRC filter: