Design Considerations for 5G mm-wave Receivers Stefan Andersson, Lars Sundström, and Sven Mattisson
Outline Introduction to 5G @ mm-waves mm-wave on-chip frequency generation mm-wave analog front-end design Noise Carrier frequency aspects Conclusions Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 2
Introduction Beam-formed transmission To enable the capacity, data rate, and coverage needed in the 5G era Why mm-wave? Large amounts of spectrum high capacity Very wide bandwidth per carrier very high data rates Beamforming with large number of antennas possible and needed due to propagation conditions Design considerations for mm-wave receivers? Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 3
Phase Noise (PN) Impact RX PN: 16QAM 1024QAM Symbol errors Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 4
PN Challenges at mm-wave Large PLL divider ratio, N, and possibly larger loop BW make XO phase noise more critical Example: using 26MHz XO @ 3 GHz: N = f 0 /f XO = 3GHz/26MHz = 41.2dB @ 30 GHz: N=f 0 /f XO = 30GHz/26MHz = 61.2dB 20dB higher XO phase noise amplification near carrier! Solution: use large f XO Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 5
Challenges at mm-wave VCO PN increases at least 6dB when f 0 doubles, (Leeson s model) Inherently 20dB worse PN when shifting from 3GHz to 30GHz oscillation frequency What makes it even worse: Q-factor of the resonator decreases when frequency increases Signal strength in the circuit decreases when frequency increases PLL performance degrades with the increased frequency Large BW (~GHz) makes the level of PN floor critical at mmw Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 6
30GHz PN Model Model is based on a prototype RX designed in 28nm FD- SOI. PoC#1 in Flex5GWare The PLL is designed for a distributed LO architecture. One PLL per transceiver/antenna Many PLLs requires low power consumption per PLL (~20mW in this case) Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 7
30GHz PN Model Model based on measurements The offset-frequency range of the measurement is 100Hz to 400MHz. Model parameters have been set such that the noise levels out at approximately -140dBc/Hz Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 8
Analog Front-End Design Dynamic range (DR) in RX limited by Front-end insertion loss LNA and ADC Typically DR LNA >>DR ADC why automatic gain control is commonly used to map the wanted signal and interference to the DR ADC Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 9
Receiver Noise Figure A simplified RX model can be derived by lumping the frontend (FE), analog/rf receiver and ADC into three cascaded blocks Noise factor (F) and noise figure (NF): Cascaded NF (Friis formula) Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 10
Simplified LNA Noise Model F 0 is the minimum low-frequency F due to additional resistive losses (R ext ) High-frequency parasitics can be accounted for by lowering f t from its intrinsic value ESD may also contribute with another 0.7dB Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 11
LNA NF Trends Recently published LNAs By inserting F 0 =1.12 (NF=0.5dB) F t =168GHz into the F min (f c ) equation we get the depicted trend curve that matches the best reported noise figures Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 12
LNAs 5 and 10 Years Out ITRS target data suggests 10x higher f t in 15 years Applying this slope we can estimate f t values 5 and 10 years from now to predict future F min (f c ) trends We can expect significant improvements in 10 years at mm-waves but very little at low GHz Higher f t will lead to lower break-down voltage... Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 13
Routing and T/R Switch Losses Layout complexity of advanced antenna system transceivers, with several on-chip RX and TX chains, incurs additional routing losses compared to traditional designs CPW loss below some 130GHz is dominated by conductor losses and f c. By assuming a 1dB/mm loss at 95GHz we can scale it to other frequencies (assuming the same metal layer thickness) When routing on a module substrate (outside the ASIC), we can expect ten times lower loss Since the FE switch consist of transistors we may apply a frequency dependence similar to the LNA, i.e. IL SW = F sw = 1 + f c /f 0, where we have used the data point IL sw =1.4dB at 50GHz Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 14
Filter Losses Filters can be implemented in several ways but due to size constraints low-order compact filters will be required Filters and wiring are dominated by conductor losses, so here we may use a f c dependence (due to the skin effect) With an IC-like technology we can assume IL filter = 2dB at 70GHz Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 15
Total Noise Figure For the wiring we may (optimistically) assume 1mm routing on chip and 20mm on module with a loss with f c carrier dependence With the above assumptions we get the following total noise figure table: Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 16
Compression Point, Gain, and Dynamic Range When comparing two designs, e.g. at 2GHz and 28GHz, the 28GHz IL will be significantly higher Maintaining the same noise factor (Fi) for the two carrier frequencies, one needs to compensate the higher 28GHz FE loss by improving the RX noise factor This can be accomplished by: Using a better LNA Relaxing the input compression point, i.e. increasing gain Increasing the DR ADC, which comes at a power consumption penalty (4x per extra bit) Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 17
Carrier Frequency Aspects Designing a receiver at 28GHz with a 1GHz signal BW leaves much less design margin than what would be the case for a 2GHz carrier with 50MHz signal BW as the IC technology speed is similar in both cases Increasing f c from 2GHz to 28GHz (>10) corresponds to designing a 2GHz RX in about 15 years old low-voltage technology, i.e. todays breakdown voltage but 15 years old f t Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 18
Conclusions Beamforming will require a substantial increase in radio chains Power, area, and volume UEs and BSs will require same high integration level LO signal generation with low phase noise is much more challenging at mm-wave Moving to mm-wave frequency RX performance will change: Expected NF increase from 5.1dB@2GHz to 9.1dB@30GHz With CMOS technology scaling the NF@30GHz will reduce from 9.1dB to 7.7dB in ten years Increasing f c from 2GHz to 28GHz (>10x) corresponds to designing a 2GHz RX in ~15 years old low-voltage technology Design Considerations for 5G mm-wave Receivers Public Ericsson AB 2017 2017-06-22 Page 19