Conclusion
LimeADPD
The LimeADPD algorithm has been implemented, verified by measured results and is capable of cancelling any distortion above system noise floor, DACs → TX → PA → Coupler → RX → ADCs.
Improvements in ACPR and EVM have been achieved in all cases as shown in Table 11.
Configuration |
Modulation |
Psat (dbM) |
Fc (GHz) |
ACPR (dBc) |
EVM (%) |
||
|---|---|---|---|---|---|---|---|
No ADPD |
With ADPD |
No ADPD |
With ADPD |
||||
Case 1 |
10MHz LTE |
19 |
0.751 |
-40.2 |
-51.8 |
3.2 |
2.2 |
Case 2 |
20MHz LTE |
19 |
0.751 |
-40.3 |
-49.6 |
3.6 |
2.2 |
Case 3 |
10MHz LTE |
39 |
0.75 |
-37.5* |
-50.5* |
3.3 |
2.4 |
Compared to the original Peak Windowing (PW) algorithm, time-multiplexing is implemented, reducing the number of utilized multipliers. The CFR block operates at 122.88 MHz, while data sample rate is 30.72 MS/s.
The architecture of CFR FIR filter is further optimized, having in mind that PWFIR coefficients are symmetrical. This reduces the number of multiplication operations, and consequentially, the number of used FPGA DSP blocks. FPGA resources are saved in this way leaving the room for some other DSP blocks to be added, DPD for example.
The novelty not seen in the published literature so far is Peak search block which is introduced in CFR preprocessing stage to find local minimum values of the signal c(n). Compared to the original PW, the difference between local minimum values of the gain correction b(n) and the clipping signal c(n) is minimized. With this circuit, the peaks of the output signal envelope are more accurately constrained to the threshold Th, resulting in lower EVM degradation.
Another important novelty here is utilization of the interpolation and decimation blocks. These are placed in front of and after the CFR block respectively. Interpolation and decimation helps in getting better EVM results for the wider modulation formats (15MHz and 20MHz LTE, for example). In other words, this approach helps the cases when the modulation edge approaches the Nyquist frequency. With this option enabled the clipping operation becomes more precise, since peaks are better seen. Adding interpolation/decimation required some modifications in FIR filter architecture and also in the method of coefficients programming.
PAPR increased as the modulation bandwidth becomes wider. A real LTE stack signal has a higher PAPR than the test model one. Both facts are well known and expected. Due to higher PAPR, the digital gain of LTE stack is backed off by 3 dB so as to avoid digital overload. Consequently, CFR threshold Th is changed from 0.75 to 0.66.
Interpolation/decimation makes CFR algorithm almost insensitive to the modulation bandwidth.
If we take 20 MHz real life LTE stack test as the target and the most challenging case, we can say that the CFR block reduced PAPR from 11.2 dB down to 8.42 dB, while degrading EVM from 0.6% to 2.3%. In other words, PAPR is reduced by 2.78 dB while EVM is degraded by only 1.7%.
ACPR is not affected at all, neither by BB modem nor CFR algorithm, thanks to digital filtering implemented by FIR blocks.
PA linearization results achieved by LimeADPD I/Q are comparable with the LimeADPD method. It is worth mentioning that beside pre-distorter, the other digital blocks are also required in transmitter paths, such as CFR and post-CFR FIR filters.
The real benefit of using LimeADPD I/Q is that the method, besides efficient PA linearization, provides RF transmitter I/Q imbalance mitigation. The nonlinearities of the PA and I/Q modulator are minimized by non-conjugate DPD block. The FIR filter, implementing linear I/Q corrector, compensates the I/Q imbalance specifically. Another advantage of using LimeADPD I/Q is that the transceiver’s static I/Q calibration procedure is not required.
LimeADPD I/Q
The LimeADPD I/Q algorithm provides low complexity in terms of reduced number of complex-valued coefficients. This is achieved by several solutions and firstly, by choosing the even order terms form for envelope function. The architecture is further simplified by selecting the FIR length N2 equal to the memory length N1. The utilization of FIR block for I/Q corrector realization additionally reduces the number of coefficients.
The total number of complex-valued coefficients of pre-distorter implemented in LimeSDR-PCIe-5G is (N1+1)×(M1+2) = (4+1)×(3+2) = 25. Decreased number of coefficients provides more savings of FPGA resources.
Measurement results demonstrate LimeADPD I/Q capabilities:
The I/Q related imbalance images are suppressed almost down to the noise floor without sacrificing the PA output power.
Although the provided results consider the test cases where the I/Q imbalance effects are present at negative image frequencies, LimeADPD I/Q also provides satisfactory results if I/Q imbalance images are present at positive frequencies.
The performance was analyzed using multi-tone signals and LTE type of waveforms.
Equaliser
The I/Q imbalance and gain error issues arise as target Local Oscillator (LO) frequency increases; they very are evident at 3.5GHz. However, the method is required also for lower target LO frequencies.
The Equaliser is implemented on LimeSDR-PCIe-5G board in LMS#2 transmit and receive paths. The transceiver LMS#2 is dedicated for 5G New Radio (NR) applications. By utilization of Equaliser, the I/Q imbalance is reduced and gain is equalized in the wide base-band bandwidth of 100MHz.
The RX and TX equaliser circuits are implemented as FIR filters. Distinct filters are applied to I and Q signal paths. The total number of FIR filters, realized for one transceiver path, is four, and, the number of real-valued coefficients of each FIR is 16.
Equaliser utilization increases the IRR by more than 15-20 dB in baseband bandwidth of 100 MHz compared to cases when only static IQ calibration is performed.