Data traffic of future 5G telecommunication systems is projected to increase 10 000-fold compared to current rates. 5G fronthaul links are therefore expected to operate in the mm-wave spectrum with some preliminary International Telecommunication Union specifications set for the 71-76 and 81-86 GHz bands. Processing 5 GHz as a single contiguous band in real-time, using existing digital signal processing (DSP) systems, is exceedingly challenging. A similar challenge exists in radio astronomy, with the Square Kilometer Array project expecting data throughput rates of 15 Tbits/s at its completion. Speed improvements on existing state-of-the-art DSPs of 2-3 orders of magnitude are therefore required to meet future demands.
One possible mitigating approach to processing wideband data in real-time is to replace some DSP blocks with analog signal processing (ASP) equivalents, since analogue devices outperform their digital counterparts in terms of cost, power consumption and the maximum attainable bandwidth. The fundamental building block of any ASP is an all-pass network of prescribed response, which can always be synthesized by cascaded first- and second-order all-pass sections (with two cascaded first-order sections being a special case of the latter). The monolithic integration of all-pass networks in commercial CMOS and BiCMOS technology nodes is a key consideration for commercial adaptation of ASPs, since it supports mass production at reduced costs and operating power requirements, making the ASP approach feasible. However, this integration has presented a number of yet unsolved challenges.
Firstly, the state-of-the-art methods for synthesizing quasi-arbitrary group delay functions using all-pass elements lack a theoretical synthesis procedure that guarantees minimum-order networks. In this work an analytically-based solution to the synthesis problem is presented that produces an all-pass network with a response approximating the required group delay to within an arbitrary minimax error. This method is shown to work for any physical realization of second-order all-pass elements, is guaranteed to converge to a global optimum solution without any choice of seed values as an input, and allows synthesis of pre-defined networks described either analytically or numerically.
Secondly, second-order all-pass networks are currently primarily implemented in off-chip planar media, which is unsuited for high volume production. Component sensitivity, process tolerances and on-chip parasitics often make proposed on-chip designs impractical. Consequently, to date, no measured results of a dispersive on-chip second-order all-pass network suitable for ASP applications (delay Q-value (QD) larger than 1) have been presented in either CMOS or BiCMOS technology nodes. In this work, the first ever on-chip CMOS second-order all-pass network is proposed with a measured QD-value larger than 1. Measurements indicate a post-tuning bandwidth of 280 MHz, peak-to-nominal delay variation of 10 ns, QD-value of 1.15 and magnitude variation of 3.1 dB. An active on-chip mm-wave second-order all-pass network is further demonstrated in a 130 nm SiGe BiCMOS technology node with a bandwidth of 40 GHz, peak-to-nominal delay of 62 ps, QD-value of 3.6 and a magnitude ripple of 1.4 dB. This is the first time that measurement results of a mm-wave bandwidth second-order all-pass network have been reported.
This work therefore presents the first step to monolithically integrating ASP solutions to conventional DSP problems, thereby enabling ultra-wideband signal processing on-chip in commercial technology nodes.