Acoustic echo control is of vital interest for hands-free operation of telecommunications equipment. An important property of an acoustic echo controller is its capability to handle double-talk and be able to operate in real time. In this paper, we propose a novel hardware architecture to support a robust adaptive algorithm in combination with a two-path model to tackle the double-talk situation. In order to achieve computational efficiency, the echo-canceller is implemented in the frequency domain and is improved by bitwidth optimisation. We show that the echo canceller is successful in handling double-talk and the sub-band implementation has improved convergence significantly. An implementation with 6 instances on a Xilinx XC4VSX55 FPGA at 165MHz can run 72 times faster than software on a 3.2GHz Pentium-4 PC.