CN101968963A - Audio signal compressing and sampling system - Google Patents

Abstract

The invention relates to an audio signal compressing and sampling system. The system comprises a frequency mixer, wherein an input end of the frequency mixer receives a random pulse signal and a to-be-modulated audio signal; an output end of the frequency mixer is connected with a reversing amplifier through a low-pass filter; and the reversing amplifier is connected with a computer through an analog/digital (A/D) converter. The system is designed according to a compressed sensing theory; and a down sampling process is performed in the way that: information of an original signal can be acquired by uniformly coating the analogue audio signal in an entire frequency domain range by modulation and sampling the demodulated signal with low sampling frequency. A part of signal is intercepted by filtering the signal with a filter with low cut-off frequency. The filtered signal comprises the information of the original signal; and then the signal is reconstructed by using a compressed sensing theory-related reconstruction algorithm. Because the sampling frequency required by the compressing and sampling system is greatly reduced, the system reduces the hardware cost of a sampling switch and the channel transmission load.

Description

The audio signal compression sampling system

Technical field

The present invention relates to a kind of audio signal compression sampling system.

Background technology

Traditional signals collecting, the theoretical foundation of encoding-decoding process are Shannon's sampling theorems, to be higher than signal highest frequency 2 overtones bands to the input analog signal sampling, according to actual needs these data are transmitted, process, are handled again.This processing gimmick defines following two point defects: one, in many practical applications because signals sampling speed must not be lower than 2 times of signal bandwidth, this makes hardware system be faced with the pressure of very big sampling rate, Nyquist sampling hardware cost costliness, obtain inefficiency, in some situation even can't realize; Its two, data storage and transmission aspect, common way are earlier to obtain data according to the Nyquist sample mode, then the data that obtain are compressed, data after will compressing are at last stored or are transmitted, and obviously, such mode causes the wasting of resources significantly.

In recent years, a kind of emerging compression sensing theory has obtained researchist's extensive concern, it is traditional data to be obtained with data compression unite two into one that this theory can be understood as, be different from the transfer encoding mode of " sampling earlier; aftertreatment " of general signal, compression sensing (Compressed Sensing, CS) according to most of signal can be on specific base this character of rarefaction representation, non-self-adapting linear projection to signal, passing through the accurate reconstruct original signal of numerical optimization problem, compression sensing theory is directly obtained a small amount of coefficient that comprises most signals in the sparse or compressible signal by a spot of linear projection at random, realizes sampling and compression function to original signal.This compression sensing mechanism has just realized the data volume compression at the end that obtains of information, has realized an innovation of signals collecting mode.Simultaneously, this theory has the character such as asymmetric of versatility, encryption, robustness, gradual, scalability, calculating.The compression sensing data obtains can regard a kind of obtaining mode of signal efficiently that is different from the Shannon sampling as, rebuilds original signal by the means of calculating from the incomplete measuring assembly of signal again.

At present, the research to the compression sensing both at home and abroad mainly concentrates in the optimization of the reconstruction algorithm that compresses the sensing rear end, and has progressively obtained perfect.Make compression sensing theory move towards application, the physics realization mechanism that the information of its front end is obtained is present problem demanding prompt solution

Summary of the invention

The object of the present invention is to provide the audio signal compression sampling system that a kind of low sample frequency, cost are low, reduce Channel Transmission pressure.

For achieving the above object, the present invention has adopted following technical scheme: a kind of audio signal compression sampling system, comprise frequency mixer, the input end of frequency mixer receives randomly pulsed phase signal and sound signal to be modulated, the output terminal of frequency mixer links to each other with sign-changing amplifier by low-pass filter, and sign-changing amplifier links to each other with computing machine by A/D converter.

As shown from the above technical solution, the present invention is according to compression sensing Design Theory, on frequency domain, be applied to equably in the whole frequency domain scope by the sound signal of modulation simulation, with lower sample frequency restituted signal is sampled, just can obtain the information of original signal, this is a down-sampled process.The wave filter lower with cutoff frequency carries out Filtering Processing to this, the part of intercept signal.Filtered signal comprises the information of original signal, and then realizes the reconstruction of signal with the theoretical relevant reconstruction algorithm of compression sensing.Because the required sample frequency of compression sampling system reduces greatly, this system has not only reduced the hardware cost of sampling switch, and has reduced Channel Transmission pressure.

Description of drawings

Fig. 1 is a circuit block diagram of the present invention;

Fig. 2 is circuit theory diagrams of the present invention.

Embodiment

A kind of audio signal compression sampling system, comprise frequency mixer 4, the input end of frequency mixer 4 receives randomly pulsed phase signal and sound signal to be modulated, and the output terminal of frequency mixer 4 links to each other with sign-changing amplifier 6 by low-pass filter 5, and sign-changing amplifier 6 links to each other with computing machine by A/D converter 7.Also comprise single-chip microcomputer 1, single-chip microcomputer 1 output terminal links to each other with the input end of activated amplifier 2, the output terminal of activated amplifier 2 links to each other with the input end of totalizer 3, the input end of totalizer 3 also inserts the offset level signal, totalizer 3 output randomly pulsed phase signals are to the input end of frequency mixer 4, as shown in Figure 1.

The present invention mainly is made up of modulation, filtering and down-sampled three parts, the sound signal of the simulation that sparse signal can will be imported on frequency domain by demodulating equipment on Fourier's base is applied in the whole frequency domain scope equably, with lower sample frequency restituted signal is sampled, just can obtain the information of original signal, this is a down-sampled process.Wave filter with cutoff frequency lower (than 2 times of much lower wave filters of the highest frequency of signal) carries out Filtering Processing, the part of intercept signal to this.Filtered signal comprises the information of original signal, and filtered signal is carried out down-sampled acquisition y[m], and then with compressing the reconstruction of the theoretical relevant reconstruction algorithm realization of sensing to x (t).

Signal is by realizing modulation with a pseudorandom PN sequence, and this sequence must one after the other between value ± 1, and frequency is far above sample frequency.Modulated process can be realized by a frequency mixer 4.In the system amplitude be the high-low level at random that directly produces by single-chip microcomputer 1 of ± 1 square-wave modulation signal at random through amplifying, adding up with offset level obtains, and realizes spread spectrum multiplying each other by sound signal to be transmitted and this random pulses.This randomly pulsed phase signal has the seemingly character of noise, and therefore the sequence that not only has been easy to generate but also can has processed and duplicate satisfies this requirement easily.Finishing modulation spectrum, to expand selected physical component be frequency mixer 4, and frequency mixer 4 is devices that are used for calculating x (t) * m (t), and m (t) signal refers in particular to the randomly pulsed phase signal that is made of ± 1 sequence.Frequency mixer 4 is realized by difference channel usually.

Usually the quantic of compression sensing can be written as:

y _j＝<f(t)，φ _j>，φ∈Φ，j∈1，…，M

Utilize the form of matrix, measuring process can be designated as:

y＝Φx＝ΦΨa

Wherein, a represents the conversion coefficient vector that m is sparse; Ψ is called as sparse base or descriptive system, and Φ is for measuring base or sensor-based system; Title M dimensional vector y is for measuring vector, and the M * N dimension matrix that claims to be made of the basic Φ of measurement is for measuring matrix.Its target be by M (M＜＜N) the inferior measurement vector y that measures measures accurate reconstruction or approximation signal x.In order to recover this signal, demoder need be carried out all

The combination of the sparse subspace of individual possibility.Therefore, Donoho has proposed based on minimizing l ₁The linear optimization model of mould, just utilize based on the base of linear programming follow the trail of (Basis Pursuit, BP) method is found the solution:

$\hat{x}=\arg \min {\left|\right|x\left|\right|}_1$ s，t y＝Φx

Wherein Φ should satisfy appearance conditions (RIP) such as restriction, promptly

The non-linear output y (t) of frequency mixer=f (t) * P (t).Can be write as:

$y_j=y\left(t\right){|}_{t=j{T}_D}={<f\left(t\right),P_c\left(t\right)>}_{t=j{T}_D}=\underset{j=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}{\&Integral;}_{(j-1)T_D}^{j{T}_D}f\left(\mathrm{\&xi;}\right)P_c\left(\mathrm{\&xi;}\right)\mathrm{d\&xi;}$

Because modulation signal P (t)=± 1, so in time domain, be the equal of that segment polarity with input signal reverses through modulation.In spread spectrum communication system, the frequency of the randomly pulsed phase signal often highest frequency than signal is high, it is wide many that its bandwidth ratio original signal is wanted, by the signal Processing ultimate principle as can be known in the time domain product of signal be equivalent to the convolution of signal in the frequency domain, principle by convolution can be known, bandwidth after both multiply each other is both bandwidth additions, thereby bandwidth is broadened, each frequency band of the signal after the modulation that the information of while input signal is also smeared.

Next be that the signal after the modulation is carried out low-pass filtering, mainly will finishing of task is exactly that the signal extension frequency spectrum that will obtain in the previous step passes through a low-pass filter 5, be equivalent to adopt a unsteady window, realize the purpose that continuous time signal elder generation block sampling is added up and sues for peace.Signal through wave filter as shown in the formula:

$y\left(t\right)={\&Integral;}_{-\&infin;}^{\&infin;}x\left(\mathrm{\&xi;}\right)P_c\left(\mathrm{\&xi;}\right){\mathrm{\&chi;}}_{[0,T_D]}(t-\mathrm{\&xi;})\mathrm{d\&xi;}=[x\left(t\right)\&CenterDot;P_c\left(t\right)]*h\left(t\right)$

Wherein:

H (t) is an ideal rectangle floating frame response, is difficult to realize that with side circuit we can introduce the integrator of an approximate ideal, and promptly low-pass filter 5 is realized.

Will finishing in the final step of task is that the signal y (t) that back obtains is transformed to a limited discrete signal y[m through over-sampling, quantification, three steps of coding].The transposition that adopts is traditional A/D converter 7.This moment, sample frequency was littler 3～5 times than the Nyquist frequency of this sound signal, will obtain y[m then] rebuild with the theoretical relevant reconstruction algorithm of compression sensing.

As Fig. 1, shown in 2, described single-chip microcomputer 1 adopts chip AT89C52, described activated amplifier 2 adopts chip U2, described totalizer 3 adopts chip U6, the 28th pin of chip AT89C52 links to each other with an end of resistance R 2, the other end of resistance R 2 links to each other with the 2nd pin of chip U2, the 3rd pin ground connection of chip U2, the 6th pin of chip U2 links to each other with an end of resistance R 13, the other end of resistance R 13 links to each other with the 2nd pin of chip U6, the offset level signal passes through the 2nd pin of resistance R 14 input chip U6, the 3rd pin ground connection of chip U6, and the 6th pin of chip U6 links to each other with the input end of frequency mixer 4.

As shown in Figure 1, 2, described low-pass filter 5 adopts chip U4, described sign-changing amplifier 6 adopts chip U5, described A/D converter 7 adopts chip U3, the 2nd pin of chip U4 links to each other with the output terminal of frequency mixer 4 by resistance R 3, the 3rd pin ground connection of chip U4, the 6th pin of chip U4 links to each other with the 2nd pin of chip U5 by resistance R 11, the 3rd pin ground connection of chip U5, the 6th pin of chip U5 links to each other with the 23rd pin of chip U3 by resistance R 5, and the 2nd, 3 pins of chip U3 link to each other with computer by serial.

As shown in Figure 1, 2, the output terminal of the output terminal of the output terminal of described totalizer 3, sound signal to be modulated, frequency mixer 4 and sign-changing amplifier 6 meets input interface A, B, C, the D of oscillograph 8 respectively.Described chip U2, U4, U5, U6 are chip AD824P, and described chip U3 is chip ATMEGA8.The 14th, 15,16,17,18,19,9,10 of described chip U3 connects A, B, C, D, E, F, G, the CP pin of charactron 9 respectively, the 6th pin of chip U3 connects 4 pin of charactron 9 by resistance R 10, the 11st pin of chip U3 connects 3 pin of charactron 9 by resistance R 9, the 12nd pin of chip U3 connects 2 pin of charactron 9 by resistance R 8, and the 13rd pin of chip U3 connects 1 pin of charactron 9 by resistance R 7.Oscillograph 8 is used for the waveform of each port is shown, charactron 9 is used to show the output after the sampling.

When work, the pulse signal of the high-low level at random of single-chip microcomputer 1 living 0 and 5V, by activated amplifier 2 its voltage is become 0 and 2V then, subtract each other by totalizer 3 direct current offset level again with+1V, just obtained amplitude and be ± randomly pulsed phase signal of 1V, utilize this pulse signal to multiply each other as the sound signal of random signal and input again and realize modulation to input signal, whole modulated process is to be realized by the frequency mixer 4 that difference channel constitutes; Next again the signal of frequency mixer 4 output is carried out filtering by a cutoff frequency far below the low-pass filter 5 of sound signal highest frequency, low-pass filter 5 usefulness one order inertia circuit are approximate to be realized, the cutoff frequency of low-pass filter 5 is lower 3～5 times than sound signal highest frequency, for the energy attenuation behind the compensation filter, filtered signal is amplified through a sign-changing amplifier 6 again, and sign-changing amplifier 6 output end signals are exactly through ovennodulation, filtered signal.

The signal of sign-changing amplifier 6 outputs is realized the A/D conversion with chip ATMEGA8, because the signal after the modulation passes through low-pass filtering again, its highest frequency is much smaller than the highest frequency of sound signal, so the A/D inversion frequency of this moment is lower 3～5 times than Nyquist sample frequency, data after will sampling are input to computing machine by computer by serial by data collecting card, it is the digital signal that comprises the sparse value of sound signal that this computer-chronograph receives signal, and computing machine utilizes the reconstruction of the theoretical relevant reconstruction algorithm realization of CS to input signal according to the digital signal that collects again.The present invention can use many data of lacking than Shannon sampling to realize reconstruction to input signal, has significantly reduced the transmission pressure of channel, has reduced the hardware cost of A/D switch.

Claims (7)

1. audio signal compression sampling system, it is characterized in that: comprise frequency mixer (4), the input end of frequency mixer (4) receives randomly pulsed phase signal and simulating signal to be modulated, the output terminal of frequency mixer (4) links to each other with sign-changing amplifier (6) by low-pass filter (5), and sign-changing amplifier (6) links to each other with computing machine by A/D converter (7).

2. audio signal compression sampling system according to claim 1, it is characterized in that: also comprise single-chip microcomputer (1), single-chip microcomputer (1) output terminal links to each other with the input end of activated amplifier (2), the output terminal of activated amplifier (2) links to each other with the input end of totalizer (3), the input end of totalizer (3) also inserts the offset level signal, and totalizer (3) output randomly pulsed phase signal is to the input end of frequency mixer (4).

3. audio signal compression sampling system according to claim 2, it is characterized in that: described single-chip microcomputer (1) adopts chip AT89C52, described activated amplifier (2) adopts chip U2, described totalizer (3) adopts chip U6, the 28th pin of chip AT89C52 links to each other with an end of resistance R 2, the other end of resistance R 2 links to each other with the 2nd pin of chip U2, the 3rd pin ground connection of chip U2, the 6th pin of chip U2 links to each other with an end of resistance R 13, the other end of resistance R 13 links to each other with the 2nd pin of chip U6, the offset level signal is by the 2nd pin of resistance R 14 input chip U6, the 3rd pin ground connection of chip U6, the 6th pin of chip U6 links to each other with the input end of frequency mixer (4).

4. audio signal compression sampling system according to claim 2, it is characterized in that: described low-pass filter (5) adopts chip U4, described sign-changing amplifier (6) adopts chip U5, described A/D converter (7) adopts chip U3, the 2nd pin of chip U4 links to each other with the output terminal of frequency mixer (4) by resistance R 3, the 3rd pin ground connection of chip U4, the 6th pin of chip U4 links to each other with the 2nd pin of chip U5 by resistance R 11, the 3rd pin ground connection of chip U5, the 6th pin of chip U5 links to each other the 2nd of chip U3 by resistance R 5 with the 23rd pin of chip U3,3 pins link to each other with computer by serial.

5. audio signal compression sampling system according to claim 1 and 2 is characterized in that: the output terminal of the output terminal of described totalizer (3), sound signal to be modulated, frequency mixer (4) and the output terminal of sign-changing amplifier (6) meet input interface A, B, C, the D of oscillograph (8) respectively.

6. according to claim 3 or 4 described audio signal compression sampling systems, it is characterized in that: described chip U2, U4, U5, U6 are chip AD824P, and described chip U3 is chip ATMEGA8.

7. audio signal compression sampling system according to claim 4, it is characterized in that: the 14th, 15,16,17,18,19,9,10 of described chip U3 connects A, B, C, D, E, F, G, the CP pin of charactron (9) respectively, the 6th pin of chip U3 connects 4 pin of charactron (9) by resistance R 10, the 11st pin of chip U3 connects 3 pin of charactron (9) by resistance R 9, the 12nd pin of chip U3 connects 2 pin of charactron (9) by resistance R 8, and the 13rd pin of chip U3 connects 1 pin of charactron (9) by resistance R 7.

Images