Add downlink PAPR measurement to eNB.

[cbalint13]

This PR introduce signal measurements for eNB in terms of quality within power dynamics.

• For now it adds PAPR (Peak to Average Power Ratio) measure routine, available as immediate command:

Available commands: 
          t: starts console trace
          m: downlink signal measurements
          q: quit srsenb
  cell_gain: set relative cell gain

• Example (by pressing m in CLI):

Cell #0 port #0 PAPR = 14.85 dB 
Cell #0 port #1 PAPR = 15.27 dB 

Future indicators for the generated signals can be added like e.g. CCDF (histogram like) to better reflect the PAPR presence.

Effective resulted PAPR for srseNB is 15dB on average, regardless PRBs and/or MIMO/SISO setup.
Experiments in lab condition (limited) with exposure to such high PAPR confirms the challenges as described below.

---

PAPR implications

Some of PAPR roles: https://www.mpdigest.com/2018/07/24/gans-role-in-5g

• PAPR is a very important indicator of radio signal when about getting it on-air with some power.
• High PAPR will give impact on final PA amplifiers design, implicit on the signal's way to the air.

PAPR at 15dB is very high for practical on-air broadcast:

• means that to get it to say 1W (30dBm) a PA of 31W(45dBm) is needed for no distorsions.
• may stretch the dynamic range for most commercial LNA, including the ones built in the SDR's rf path.
• commercial PA solutions for LTE signals can handle maximum of 9dB PAPR (e.g. NXP's AirFast LDMOS series)
• advanced PA design (like Doherty) summ two/multiple PA RF paths (AB carriers, C peakers), but still not >10dB of PAPR.

Various OFDM PAPR limits: https://www.mpdigest.com/wp-content/uploads/2018/07/Figure01.jpg

OFDM	dB
3G-WCDMA	3.5dB
3G-HSUPA	6.5dB
LTE	8.5dB

---

Suitable PAPR reduction

A good presentation for PAPR efforts can be found here available on public github

• The classical Clipping is the simplest implementation but it degrades the BER.

• The mentioned SelectiveMapping with this snippet fits srseNB, at cost of random phase elections, zero BER impact.

• PAPRnet yields state-of-art, at cost of intense tensor-operators, it make sense if srseNB reach to GPU platforms in future.

---

@andrepuschmann @suttonpd
Please help with the review, looking forward for suggestions & enhanchments.

Thank You !
~cristian.

[ismagom]

This looks amazing, thanks for this great contribution!

I am not very familiar with PAPR reduction techniques, but looking at SelectiveMapping it looks like it requires the receiver to be aware of the changes, same as PAPRnet. Do you think these can be used with COTS UE?

I wonder if you are aware of any "simple" PAPR technique we could use that is compatible with LTE standard? other than clipping of course

Thanks again!

[cbalint13]

This looks amazing, thanks for this great contribution!

I am not very familiar with PAPR reduction techniques, but looking at SelectiveMapping it looks like it requires the receiver to be aware of the changes, same as PAPRnet. Do you think these can be used with COTS UE?

Some PAPR reductions like ToneReservation requires receiver side awareness:

• https://wiki.gnuradio.org/index.php/Tone_Reservation_PAPR

SelectiveMapping does not require receiver side (UE) awareness:

• SM "elects" initial phase from random mappings in [1,-1, j, -j ] for a lower PAPR
• SM finds by "election" an equivalent OFDM version of the same signal.

SM looks a good candidate, would be a way (IMO, easiest so far) to have it in srsRAN (see sample code template).
Not looked thoroughly at way how to add it, might be some catches making it unsuitable (need dwelve into more deeply).

The 9dB PAPR would be the normal (bellow would be excellent) for srsRAN to achieve.

I wonder if you are aware of any "simple" PAPR technique we could use that is compatible with LTE standard? other than clipping of course

PAPRnet is fabulous one, but CNN tensor operator are quite intense. I think such approaches would be possible in the future context of #587 (see my latest note there) but trained to reduce PAPR without receiver side awareness.

Thanks again!

With pleasure !
~cristian.

[cbalint13]

• As update and correction to above info all SM (selective mapping) require some sort of UE awareness:

  a. SM is most effective when re-mapping info is transmitted on side-channel to UE
  b. SM without side-channel but UE should still be aware of possible OFDM permutations
  

• Other reduction models beyond clipping are it's windowed variants (also RTL friendly): PW-CFR & PC-CFR

• The ML based approach like PAPRnet have to be trained without any receiver side awareness.

[CLAassistant]

All committers have signed the CLA.