DSP_DPD

• offline DPD study
• DPD model reference from https://github.com/ctarver/ILA-DPD

1. x: signal pa in
2. generate pa model(Memoryless):

• IIP3dBm = 40
• AMPMdeg = 1
• LinearGaindB = 10
• PowerUpperLimit = 45
• Ripple = 0

3. y: signal pa out and aclr polot, evm_y :6.235%

4. Learning parameters:

• dpdparams.order_poly = 3+2*1
• dpdparams.depth_memory = 1+2*1
• dpdparams.depth_lag = 2
• dpdparams.depth_memory_lag = 2
• dpdparams.order_poly_lag = 2
• dpdparams.Niterations = 20
• dpdparams.learning_rate = 0.8
• dpdparams.learning_method = []
• dpdparams.flag_even_order_poly = 1
• dpdparams.flag_conj = 0; % Conjugate branch. Currently only set up for MP (lag = 0)
• dpdparams.flag_dc_term = 0; % Adds an additional term for DC
• dpdparams.flag_LS_exclude_zero_second = 0
• dpdparams.modelfit = 'WIN' % 'GMP'/'HAM'/'WIN'
• dpdparams.learning_arc = 'DLA';

5. DPD learning and curvefit iteration

6. ACLR upgrade from 32.77 to 46.92, evm upgrade from 6.235% to 0.8789%

7. PAR results
   

8. Model/ Architecture /Iterations comparsion
   

9. 2C signal

	1C Pwr(dBm)	2C Pwr(dBm)
x	14.9	11.89, 11.89
y	34.8	31.80, 31.80
y+DPD	35.0	31.92, 31.92
	1C ACLR(dBm)	2C ACLR(dBm), (L1,C,U1)
x	55.2, 55.1	54.6, 52.1, 54.6
y	32.7, 32.7	32.6, 29.7, 32.6
y+DPD	47.9, 46.9	50.4, 48.4, 50.5
	1C PAR (dB)	2C PAR (dB)
x	10.9	13.9
y	10.6	13.8
y+DPD	11.1	13.3

• intermodulation improvement, 2C
• PAR, 2C
  

10. Add Ripple to ORX full bandwidth

• paRipple = 10 %% 2021-10-22, Add Ripple to ORX
• DPD result: evm:0.7091%, ACLR:49dB, the Ripple will be optimizied by learning

pa Ripple (dB)	ACLR L(dB)	ACLR U(dB)
0	51.7	51.5
10	49.25	49.9

2021-11-09,
11. Add ORX SNR parameter, sweep SNRdB 50:-10:10

• dpdparams.ORX_SNRdB = 10
• The ORX SNR decrease to 10dB, that impact the DPD ACLR results about 3dB.

ORXSNR (dB)	ACLR (dB)	EVM (%)	Inband Pwr (dBm)
50	51.7, 51.5	0.38	34.95
40	51.7, 51.5	0.38	34.95
30	51.6, 51.4	0.38	34.96
20	51.0, 51.2	0.40	35.11
10	48.1, 48.4	0.68	36.32

12. Add ORX Ripple(Fullband 122.88MHz) parameter, sweep RippledB 0:2:10

• The ACLR will be unbalanced and results worse about 4dB at Low freqs.

ORX Ripple (dB)	ACLR L(dB)	ACLR U(dB)	EVM (%)	diff ACLR (dB)
0	51.7	51.5	0.38	0.2
2	51.0	51.9	1.06	0.9
4	50.3	52.1	2.01	1.8
6	49.5	52.0	2.99	2.5
8	48.7	51.7	3.97	3
10	47.8	51.3	4.96	3.5

going to do...

13. compare the DPD ACLR performance between differenet ACLR source ?

ACLR(dB)	w/o AWGN	w/ AWGN 40dB
x	55	47
y	32	32
y+DPD	51	46

14. compare the ccdf 0.01% between different modulation type?

• PA settings:

Parameters	paIIP3dBm	paAMPMdeg	paLinearGaindB	paPowerUpperLimit
	31dBm	1deg	20dB	35dBm

• x, y, y+DPD, PAR vs ACLR

	w/o CFR _ PAR(dB)	w/ CFR _ PAR(dB)	w/o CFR _ Pwr(dBm)	w/ CFR _ Pwr(dBm)	w/o CFR _ ACLR(dB)	w/ CFR _ ACLR(dB)
x	10.9	7.5	14.99	14.96	55.22/ 55.17	55.06/ 55.02
y	8.1	6.6	34.54	34.52	31.46/ 31.44	31.63/ 31.65
y+DPD	8.1	7.4	34.54	34.55	44.35/ 44.46	50.30/ 50.21

• y+DPD, EVM (compare to x, what is the CFR contribute to EVM ?)

ACLR(dB)	w/o CFR _ EVM	w/ CFR _ EVM
y+DPD	1.05	0.47

15. compare the DPD performance between different source IBW ?

16. compare the input source x' = x add dpd coeffiecents at time and freqency domain?
    

17. Add PA memory model and run DPD 17b. Compare PA memory model of the AMAM, AMPM of w/i and w/o DPD
    18. What is the CFR contribute to EVM ?

18. Releationship between OIP3dBm and ACLR, w/i and w/o DPD
    20. CCDF at 7.5, 8.5, 9.5 condidtions, w/i and w/o DPD