1 (12) Dirty-RF Theme Some Radio Implementation Challenges in 3G-LTE Context Dr. Mikko Valkama Tampere University of Technology Institute of Communications Engineering mikko.e.valkama@tut.fi
2 (21) General Dirty-RF Paradigm General challenge: Low-cost, low-power, small-size yet flexible radio implementations (TX and RX) for future wireless systems. Cost and size of individual radios (are or should be) going down especially important in multiantenna (MIMO) systems with multiple parallel radios simplified radio architectures and low-cost analog electronics, especially for the RF parts With simplified analog RF front-ends, several RF impairments or non-idealities <=> dirty-rf theme getting more important also when the waveform structure gets more complex (higher-order modulations, etc.), e.g. 3G-LTE downlink with 64QAM-OFDMA or 3G-LTE uplink with 16QAM-SC-FDMA Some important example problems: I/Q imbalance, PA nonlinearities, oscillator phase noise, timing jitter in IF/RF sampling, mixer and LNA nonlinearities in receivers, etc.
3 (21) General Dirty-RF Paradigm (cont d) Instead of increasing the design efforts and cost of the used analog electronics, use sophisticated DSP for restoring and enhancing the quality of the RF parts; DSP-based mitigation of RF impairments LPF A/D I I RF COMP. CHAN. EQ. BPF I/Q LO SELECTIVITY DEMOD. DETECTION DECODING LPF A/D Q SYNCH. Q SYNCH. ANALOG FRONT-END DIGITAL FRONT-END DIGITAL BASEBAND In general, the dirty-rf theme calls for expertise on both sides of the A/D and D/A interfaces; high synergy by merging the radio engineering and baseband DSP communities and knowledge!
4 (21) General Dirty-RF Paradigm (cont d) In the earlier systems, with dedicated hardware and narrowband low-order modulated waveforms, these RF impairments did not actually play that big a role. But the story is really totally different in the emerging future systems increasing bandwidths, more and more complex multicarrier type waveforms (with increased sensitivity to all distortion and interference) relative levels of RF impairments increasing due to simplified RF parts and electronics, and also due to flexibility and reconfigurability requirements In general, seen to play a major role in the future wireless evolution!
One Important Practical Example Problem: I/Q Imbalance and Mirror-Frequency Interference 5 (21) I/Q imbalance: unintentional amplitude and phase errors between transceiver I and Q signal branches Results in mirror-frequency interference practical mirror-frequency attenuation in the order of 20-40dB, depending on the quality (amplitude and phase error levels) of the analog front-end the role depends heavily on the applied transceiver architecture (zero-if, low- IF, etc.) and on the used communications waveforms most critical in wideband multicarrier IF transmitters and receivers, in which the mirror-frequency band is another signal band (high dynamic range) high modulation order also increases sensitivity Illustrated in the following figures, for both TX and RX
I/Q Imbalance and Mirror-Frequency Interference 6 (21) Mirror-frequency interference due to I/Q imbalances on TX side Zero-IF IQ I/Q LO RF 0 f I D/A LPF 0 f RF = f LO f Low-IF IQ RF RF 0 f IF f Q D/A LPF 0 f RF 2f f LO IF f RF f Interference to other signal bands (low-if) or within the spectrum itself (zero-if)
I/Q Imbalance and Mirror-Frequency Interference 7 (21) 3G-LTE TX scenarios: for fast frequency domain scheduling and frequency hopping capabilities, the focus in 3G-LTE uplink TX (mobile) is most likely on the latter IF transmitter case => tuning within the total system/operator band (5-20MHz) on the digital side => big challenges to obtain sufficient attenuation for the mirrorfrequencies, at least using purely analog techniques! the individual mobile signals in the corresponding downlink TX (basestation) also (anyway) follow the IF transmitter model => again big challenges additional mirror-frequency attenuation using DSP-based calibration techniques notice also that due to limited power-control, the mobile transmitter case is more challening (from the TX side point of view)
I/Q Imbalance and Mirror-Frequency Interference 8 (21) Mirror-frequency interference due to I/Q imbalances on RX side I/Q LO RF LPF A/D I IQ RF 0 f LO f RF,i f LPF A/D Q 0 f IF,i f Interference between different carriers or parts of the used radio spectrum
I/Q Imbalance and Mirror-Frequency Interference 9 (21) 3G-LTE RX scenarios: (again) for fast frequency domain scheduling and frequency hopping capabilities, the overall system/operator band (5-20MHz) is mostly likely I/Q downconverted as a whole (wideband direct-conversion / low-if radio architecture) => this applies most likely also to mobile receivers as well (=> base-station is, of course, demodulating all the carriers anyway) so in both mobile as well as base-station receivers, the individual mobile signals follow essentially the IF model => very big challenges to obtain sufficient mirror-frequency attenuation! => can and should be enhanced using proper DSP, I/Q imbalance compensation theme
Example case study #1: TX I/Q Calibration Using Pre-Distortion and Feedback 10 (21) Mobile TX
Example case study #1: TX I/Q Calibration Using Pre-Distortion and Feedback 11 (21) Proper pre-distortion of the digital signal such that generated RF signal is essentially free from mirror-frequency interference Pre-distortion coefficients estimated based on the statistics between the feedback signal and the known TX data real down-conversion based feedback to IF to avoid excess I/Q errors can handle, by design, also frequency-dependent I/Q imbalances No technical details or math here, simply iillustrated using a simple example this example focuses on IF transmitter, but similar principles apply also in case of zero-if transmitter and/or wideband multicarrier TX
Example case study #1: TX I/Q Calibration Using Pre-Distortion and Feedback 12 (21) Example: 3G-LTE SC-FDMA low-if transmitter with 16-QAM data, roughly 2 MHz mobile bandwidth, IF frequency 3 MHz. Frequency-selective I/Q mismatches (analog front-end) varying within the overall signal band results in frequency-selective analog front-end mirror-frequency attenuation varying between ~25dB... ~40dB Estimator block-size 100,000 and pre-distortion filter length = 3. Figures below show the average obtained IRR (in many independent realizations) together with 10 individual realizations exceptionally good performance, stemming partially from the analytic nature of the overall waveform (signal energy only at the other side of the spectrum)
Example case study #1: TX I/Q Calibration Using Pre-Distortion and Feedback 13 (21) 90 Average IRR vs. Frequency 100 10 IRR Realizations vs. Frequency 80 90 IRR [db] 70 60 50 40 30 20-1 -0.5 0 0.5 1 Normalized frequency ω/π IRR [db] 80 70 60 50 40 30 20-1 -0.5 0 0.5 1 Normalized frequency ω/π Analog front-end in dark-dashed, overall (calibrated) in colors-solid
14 (21) Example case study #2: RX I/Q Compensation Using the Rich Statistics of the Received Complex Signal Contrary to TX, receiver front-end signals pretty challenging for estimation purposes, due to channel noise (low SNR s), multipath, (mis)synchronization, other interferences, etc. Here we focus on purely blind signal processing utilizing the rich statistical nature of the received complex signal under RX I/Q imbalance (no explicit waveform structure assumed) To put it in short: The ordinary and complementary (pseudo) correlation functions of the received signal offer sufficient information to extract I/Q compensation parameters can be estimated using sample statistics directly from the received complex signal (overall down-converted signal), rather independently of the exact waveform structure (type of modulation, etc.) unaffected by noise, multipath, synhronization, etc. can again handle also the challenging case of frequency-selective I/Q imbalances
Example case study #2: RX I/Q Compensation Using the Rich Statistics of the Received Complex Signal 15 (21) AGC LPF A/D I BPF RF LNA I/Q LO I+ jq LPF AGC A/D Q (.)* w( t) Post-Processing Compensator IQ IQ 0 f IF,i f 0 f IF,i f
16 (21) Example case study #2: RX I/Q Compensation Using the Rich Statistics of the Received Complex Signal Example: Base-station receiver with 4 rather wideband mobiles. 10MHz 3G-LTE uplink mode assumed and here used by 4 mobiles 16-QAM SC-FDMA waveforms for each mobile individual mobile bandwidths of {4.5, 2.16, 1.26, 1.08}MHz corresponding RF power levels at BS input are {0, 10, 15, 20}dB Individual multipath fading channels plus noise for each mobile drawn from the Extended Vehicular A power-delay profile the velocities of the four mobiles are {30, 3, 120, 240}km/h and the RF carrier range is 2GHz RX frequency offset (CFO) also included Frequency-selective I/Q mismatches are assumed for the basestation RX front-end, varying smoothly between 30-40dB. A three-tap post-processing compensation filter is used and a block of 100,000 received samples is used for sample statistics evaluations.
Example case study #2: RX I/Q Compensation Using the Rich Statistics of the Received Complex Signal 17 (21) 80 70 Mirror Frequency Attenuation vs. Frequency Compensated, Average Compensated, Realizations Analog Front End 60 Attenuation [db] 50 40 30 MS#1 MS#2 MS#3 MS#4 20 0 db 10 db 15 db 20 db 10 6 4 2 0 2 4 6 Relative Frequency [MHz]
18 (21) Example case study #2: RX I/Q Compensation Using the Rich Statistics of the Received Complex Signal Clearly, good compensation performance is obtained, especially at the most weak signal band (here the most wideband mobile) automatically most mirror-frequency attenuation to those bands really needing it Next, laboratory radio signal measurements are carried out for further demonstration and verifications similar waveforms (4 SC-FDMA mobiles, etc.), state-of-the-art receiver RF-IC chip, high-quality laboratory signal generators, etc. example obtained results illustrated in the following, in terms of the detection error rate (BER/SER) when detecting the most wideband mobile signal in the base-station
Example case study #2: RX I/Q Compensation Using the Rich Statistics of the Received Complex Signal 19 (21) 50 Spectrum of the measured composite SC FDMA signal 40 30 MS 4 MS 3 MS 2 20 MS 1 Amplitude [db] 10 0 10 20 30 40 50 8 6 4 2 0 2 4 6 8 Frequency [MHz]
Example case study #2: RX I/Q Compensation Using the Rich Statistics of the Received Complex Signal 20 (21) 10 0 Measured localized SC FDMA waveforms, 16 QAM 10 1 SER 10 2 10 3 Uncompensated Iterative, N w =1 Iterative, N =3 w Block, N =1 w Block, N =3 w 10 4 No imbalance (simul.) 5 8 11 14 17 20 23 Average SNR [db]
21 (21) RF-DSP Research Group at TUT, Finland Pioneering work in the dirty-rf field since 1999, focus on understanding and mitigation of the various analog RF impairment effects using DSP. Altogether 7 PhD/MSc researchers at the moment, led and coordinated by Dr. Mikko Valkama. Several research projects focusing on the dirty-rf topics, funded by, e.g., the Academy of Finland and Nokia-Siemens Networks. Happy to make new openings and establish new research cooperation... Contact: Dr. Mikko Valkama Tampere Univ. Technology, P.O.Box 553, Tampere, Finland Email: mikko.e.valkama@tut.fi, Tel: +358-40-8490-756