V.S. Efimenko - Ph. D, professor Moscow branch of OJSC "VNIIR-Progress" (part of the holding "ABS Electro"). E-mail: vefimenko@abselectro.com
A.V. Pastukhov - Ph.D. in Technical Sciences, Moscow branch of OJSC "VNIIR-Progress" (part of the holding "ABS Electro"). E-mail: apastukhov@abselectro.com
D.N. Voronchikhin - Moscow branch of OJSC "VNIIR-Progress" (part of the holding "ABS Electro"). E-mail: dvoronchikhin@abselectro.com

One of the most important problems is improving of radio systems jam-proof. The key method of its solution is space-time processing. Optimal spatial filters (beamformers) performs 2 main operations: jam rejection (i.e. form nulls to the jamming direction in antenna pattern) and focusing to the signal (i.e. steer pattern beams to the signal direction). Jam rejection block can be noted as a multichannel canceller (MCC). MCC extraction is connected with simultaneous multiple legitimate signals receiving that only derives in multiplication of focusing blocks for different signal directions. Note that MCC doesn-t depend on the signals and may be implemented as loose module (e.g. IP-core). Beamformer implementation issues are connected with the online digital implementation of the matrix operations. Depending on hardware equipment (processors or FPGAs), it-s reasonable to use different methods to get "pure" MCC output which is focused in compliance with signal direction. Methods are one with jamming covariance matrix deriving and one without it. For GNSS signal reception, beamformer implementation has not very severe hardware productivity requirements. That allows to carry it out in to modifications: 1) with matrix accumulation and spatial processing using FPGA and matrix inversion using processor; 2) using "saving" systolic structures of QR decomposition and forward and backward substitutions. For some communication systems, implementation is similar to GNSS beamformer. Nevertheless, it has critical conditions of less than 10..15 µs observation and processing time. FPGA only implementation is carried out using QR decomposition and back substitution systolic blocks without covariance matrix deriving.