A FSK Demodulator Comparison for Ultra-low Power, Low Data-rate Wireless Links in ISM Bands Emanuele Lopelli ∗ J.D. Van der Tang ∗ A.H.M. van Roermund ∗ Abstract — Personal communications require wire- less nodes, which can transmit and receive reliably data under huge power constraints. Higher level of integration and reduction of power consumption can be achieved using a zero-IF architecture together with a wideband BFSK modulation scheme. Un- fortunately FSK techniques performances degrade sharply in the presence of frequency offset. In this paper a comparison between potentially low power BFSK architectures is presented based on high level models. In depth analysis of four potentially low power demodulators shows that the architecture, which can assure rejection of large static offset with minimum increment in hardware complexity is the ST-DFT based demodulator. This will allow great reduction in power consumption avoiding acquisition and tracking of the offset at the receiver side. 1 INTRODUCTION The market for wireless connectivity has been grow- ing rapidly in the past few years. A number of standards have been developed to satisfy the re- quirements of various parts of this market. These systems have unique characteristics and engineers will face new implementation challenges. The most severe design constraint is the limited amount of energy available for a wireless node. Beside this challenge, low cost and small size are two impor- tant features for a product that wants to reach the largest part of the consumer market. Hardware minimization can be achieved by using a zero IF architecture, which eliminates image-reject filter and other IF components, enabling a mono- lithic transceiver. In spite of its lower hardware complexity, the zero-IF architecture suffers from some special problems like DC offset and 1/f flicker noise coming from CMOS transistors. Among the various ways to cope with such problems, the most straightforward way is to remove the low-frequency portion of the downconverted signal’s spectrum by a highpass filter (HPF). However to avoid to cut a large portion of the signal energy, and to use sim- ple HPF filter it is important that the spectrum of the signal is placed sufficiently far away from DC. For these reasons a non-coherent wideband bi- nary frequency-shift-keying (BFSK) is a simple and robust form of digital modulation when the band- ∗ Technical University of Eindhoven, The Nether- lands, e-mail: [E.Lopelli, J. D. v.d.Tang, A. H. M. v.Roermund]@tue.nl, tel.: +31 40 2472666. width requirements are not overly stringent like in transceivers for low data-rate applications. Unfortunately, FSK shows limits when the fre- quency offset is nonzero, which is the normal case in practical applications due to imperfections in the local oscillator. Therefore if the demodulator can- not cope with this offset, the frequency instability should be corrected with an Automatic Frequency Control (AFC) system, before the signal can be cor- rectly demodulated. This will increase the hard- ware complexity at the receiver side and therefore the overall power consumption. The objective of this paper is to study the effect of frequency offset on different FSK demodulators un- der additive white Gaussian noise channel (AWGN) using simulations. To do so, four demodulators are throughly analyzed in the reminder of this paper and developed theory verified against simulations. 2 BFSK DEMODULATION Among all the demodulator architectures, four of them, suited for ultra-low power implementation, have been chosen and their performances have been simulated using Simulink models under different frequency offsets in a AWGN channel. These four type of demodulator are: • Arctan-differentiated demodulator (ADM) [1] • Correlation demodulator [2] • Digital cross-differentiate multiply demodula- tor (DCDM) [3] • Short-time DFT (ST-DFT) demodulator [4] Another widely used digital demodulator is the zero-crossing demodulator. Unfortunately in [5] it has been shown that when the frequency offset is equal to the 6% of the data rate, the SNR degradation reaches 4.8 dB. In the field of low data-rate applications, this will limit the maximum uncorrected frequency offset, without significant degradation in the BER performances, to a few hundreds of Hertz, which is quite difficult to achieve in practice without an AFC system. 2.1 Arctan-Differentiated Demodulator In this demodulator topology, the FM discriminator is followed by an Integration-and-Damp (IaD) cir-