Wireless noise communication system based on the transmitted reference technique (TRT) is proposed for a covert data transmission through additive white Gaussian noise (AGWN) channel. Noise waveforms with the time windows are formed for data transmission using (3.1–4.1) GHz band-pass filters with symmetrical finite impulse response (FIR). FIR filter design includes the specification of the rectangular frequency response and the selection of appropriate window functions, which satisfy pass-band and attenuation specifications. Spread spectrum noise communications apply the time diversity between the noise reference and informative noise carriers delayed at the time interval T = 6 ns exceeding the noise coherent time τc = 1 ns. The delayed noise carriers are multiplied by antipodal binary symbols bl = ±1 at the same rate to informative data stream. The delivered noise reference is transmitted through wireless channel simultaneously with delayed noise waveforms contained informative components. Spectrum modulation of transmitted waveforms is performed by means of linear superposition between the noise reference and informative noise carriers delayed at time interval T. Spectral power density of result noise signals is modulated by antipodal harmonic functions with the period in inverse proportion to relative time delay Т. The coherent convolution of continuous noise signals is produced by the correlation receiver during every informative symbol interval. The correlation time delay in the receiver channel corresponds to the diversity time delay Т of informative noise signals. The correlation estimation problem appears in the case of data transmitting on the base of continuous noise carriers. Correlation estimations are statistical evaluated for total noise signals propagating throw AGWN channel. Intersystem jamming is excited at the output of correlation receiver even if noise informative carriers are transmitted over a wireless channel without thermal noise. Autocorrelation receiver output is randomly deviated near the average value according to informative data rate. Window method is proposed for a digital compensation of random distortions in correlation estimations. It is shown, that utilizing continuously noise signals with the time windows permits to decrease randomly fluctuations of correlation estimations.