CN106027438A - Anti-aliasing modulation demodulator for analog-digital hybrid amplitude modulation broadcasting system - Google Patents

Abstract

The invention discloses an anti-aliasing modulation demodulator for an analog-digital hybrid amplitude modulation broadcasting system. According to the demodulator, aiming at a MPPSK (M-ary Position Phase Shift Keying) digital modulation signal separated from a MPPSK/DSB-AM (M-ary Position Phase Shift Keying/carrier-reserved Double Side Band-Amplitude Modulation) hybrid modulation system received signal, DL (Deep Learning) training is carried out by utilizing an SAE (Sparse Autoencoder) neural network, so that the trained DL-SAE neural network can carry out classification on an MPPSK received signal sample which generates interference between strong codes due to bandwidth limitation of a sending end, and thus, compared with demodulation bit error rates of a conventional amplitude integral judgment demodulator and matched filtering judgment demodulator, a demodulation bit error rate of the anti-aliasing modulation demodulator is reduced by at least one magnitude order, and simultaneous and hybrid transmission of an analog audio and high-speed data in a 9kHz amplitude modulation broadcasting channel is ensured.

Description

A kind of anti-aliasing modem for modulus mixing AM-broadcasting system

Technical field

The present invention relates to exist the digital communication system of intersymbol interference, particularly relate to exist the compatible amplitude modulation broadcasting of intersymbol interference The judgement demodulation problem of modulated signal of hybrid multiplex modulation system, belong to digital communication and Nonlinear harmonic oscillator field.

Background technology

Traditional analog AM (AM) is broadcasted as a kind of ancient communication system, has continued to use last 100 years, so far Cannot meet the most far away the demand of people, digital radio becomes the inexorable trend of its development.How to make full use of new technique Realize effective covering of broadcast program, it is achieved analogue audio frequency broadcast needs solution badly to the smooth transition of digital audio broadcasting；Under Face provides several solutions present in prior art.

1.MPPSK transmission system

In order to make full use of existing high-power simulation AM broadcast transmitting apparatus and radio network, it should compatible existing guarantor Stay simulation amplitude modulation double side band (DSB-AM) the transmission system of carrier wave；And DSB-AM signal is not contained any information by one Sinusoidal carrier and upper and lower two containing identical modulation intelligence simulation sidebands composition, therefore to realize simulation with numeral system Same broadcasting of system, it is necessary to utilize the carrier wave of DSB-AM signal to carry out digital modulation: if the power spectrum of digital modulation information (PSD) sideband in 9kHz (centered by carrier frequency ± 4.5kHz) bandwidth of simulation main signal at least below carrier wave 50dB, Can realize with broadcasting.We carry the carrier wave of the lowest PSD sideband digital information referred to as " digital carrier " this, and In order to produce " digital carrier ", it is necessary to consider to retain the digital modulation technique of sinusoidal carrier component.

Code element " 0 " and " 1 " of traditional binary phase shift keying (BPSK) are anti-phase, completely inhibit carrier wave, though having well Demodulation performance, but the availability of frequency spectrum is only 1bps/Hz in theory.Chinese patent ZL200710025202.1 (multielement positional Phase-shift keying (PSK) modulation and demodulation method) the phase-shift keying (PSK) modulation of multielement positional that proposes is then that a kind of modulating range is asymmetrical Phasing technique, not only remains carrier wave, and relies on multi-system modulation to improve spectrum efficiency, and this patent is to original The a kind of of MPPSK modulation simplifies and improves, and its expression formula is as follows:

Wherein: k=0,1 ..., M-1 (M ＞ 2) is the information symbol sent, and T is code-element period；ω_cFor carrier wave angle Frequency, T_c=2 π/ω_cFor carrier cycle；0≤r_g＜ 1 protects the Separation control factor for symbol, by M, K, N and r_gConstitute Change signal bandwidth, efficiency of transmission and " modulation parameter " of demodulation performance.

The more important thing is, analyze and show to work as r_gWhen=0, if met:

N=M K (2)

Then the line spectrum in MPPSK signal PSD is completely eliminated, for simulated audio signal and the neighbouring frequency of same channel The interference of other broadcast station signal in road, can be lower.

2.MPPSK/DSB-AM hybrid multiplex modulation system

Modulate based on digital carrier thought and MPPSK, Chinese patent application 201310464224.3 (a kind of compatible medium wave The hybrid multiplex modulation system of analog AM broadcast) propose a kind of MPPSK/DSB-AM hybrid multiplex modulation system.This compound tune System processed, on the basis of keeping former DSB-AM broadcast singal system constant, uses MPPSK digital carrier just replacing String carrier wave removes to carry the amplitude modulation(PAM) of analog broadcast signal.The amplitude information of modulated signal carries audio signal, and phase place Information then contains digital signal, thus in existing 9kHz medium wave bandwidth, transmits simulation and numeral two-way simultaneously Signal, substantially increases the availability of frequency spectrum.

MPPSK/DSB-AM hybrid multiplex modulation system uses traditional coherent demodulation for digital information: first to receiving And carry out coherent detection through the intermediate frequency multiplex modulated signal of down coversion and (utilize from receiving the same frequency homophase extracted signal Coherent carrier, and receive after signal multiplication low-pass filtering again) obtain baseband signal, the then location of lock-out pulse in place Under the baseband signal samples value in a code-element period carried out matched filtering judgement or the judgement of amplitude integration, can demodulate Digital modulation information.But, in order to the PSD sideband making MPPSK/DSB-AM complex modulated launch signal meets wireless The regulation of electricity administration section, reduces the damage for simulated audio signal demodulation tonequality, it is necessary to adjust MPPSK simultaneously The PSD sideband that signal processed itself is the lowest applies further spectral mask and limits, and this most greatly have impact on The transmission performance of MPPSK digital modulation signals, introduces serious intersymbol interference (ISI), even if in the highest signal to noise ratio Under, the above-mentioned traditional demodulation method demodulation bit error rate when being not added with chnnel coding the most only tends to 1% magnitude, it is necessary to invention is more Good demodulation method is to improve the demodulation performance for MPPSK/DSB-AM multiplex modulated signal.

3. degree of depth study-sparse own coding (DL-SAE)

(1) degree of depth study (DL:Deep learning) starts from University of Toronto professor Geoffrey in 2006 The paper about many hidden layers deep neural network that Hinton delivers on " science " magazine, it is machine learning One branch, is mainly characterized by obtaining the expression for initial data difference level of abstraction by multi-level study, enters And improve the accuracy of the tasks such as classification and prediction.Such as there is a collection of input I (signal as gathered under a collection of varying environment), Assume that we devise system S of a n-layer, by the parameter in adjustment system so that its output remains input I, The most just can automatically derive a series of level characteristics S of input I₁,S₂,…,S_n.Therefore, be different from traditional support to The shallow-layer learning algorithms such as amount machine (SVM), DL is without relying on artificial experience sample drawn feature, but is had by structure There are the machine learning model of a lot of hidden layer and the training data (i.e. utilizing " big data ") of magnanimity, automatically learn more useful Feature, thus finally promote classification or prediction accuracy.

(2) sparse own coding (SAE:Sparse Autoencoder) network is the one of artificial neural network.One Own coding (AE) the network only one of which hidden layer of monolayer, the output of its target is equal to input.AE can obtain input data Key character also reconstructs input signal.SAE, on the basis of AE, adds the openness restriction of network, i.e. major part The value of hidden node is 0, and only minority is non-zero.Owing to the theoretical output valve of SAE is the eigenvalue x of input itself, make The hidden layer obtaining SAE network can replace the feature of input well.Fig. 1 is a SAE network example, and its input is The activation amount of each neuron is a_i, i=1,2 ..., m, then:

A (X)=f (W₁X+b₁) (3)

Wherein f (Z)=1/ (1+exp (-Z)) is non-linear activation primitive, a (X) ∈ R^mIt is neuronal activation vector, W₁∈ m × n is weight matrix, b₁∈R^mIt it is offset vector.Network is output asWhereinIt is output valve, W₂∈ n × m is weight matrix, b₂∈RⁿAlso it is an offset vector.

A given training sample set { (x⁽¹⁾,y⁽¹⁾),(x⁽²⁾,y⁽²⁾),…,(x^(m),y^(m)), it comprises m sample.Can use Back-propagation algorithm minimizes reconstructed errorAdjust weights W₁、W₂And b₁、b₂.When network does not increases During addition of constraints condition, cost function is:

$J(W,b)=\[\frac{1}{m}\underset{i=1}{\overset{m}{\Σ}}\left(\frac{1}{2}\left|\right|h_{W,b}\left(x^i-y^i\right)|{|}^2\right)\]+\frac{\mathrm{\λ}}{2}\underset{l=1}{\overset{n_t-1}{\Σ}}\underset{i=1}{\overset{s_t}{\Σ}}\underset{j=1}{\overset{s_l+1}{\Σ}}{\left({W^{\left(l\right)}}_{ji}\right)}^2---\left(4\right)$

Wherein, Section 1 is mean square deviation item；Section 2 is regularization term (being also weight attenuation term), it is therefore an objective to reduce power The amplitude of weight, prevents overfitting.Weight attenuation parameter λ is the relative importance of two in control formula.When to hidden layer When neuron applies openness restriction, cost function can be expressed as:

$J_{sparse}(W,b)=J(W,b)+\mathrm{\β}\underset{j=1}{\overset{s_2}{\Σ}}KL\left(\mathrm{\ρ}\right|\left|{\widehat{\mathrm{\ρ}}}_j\right)---\left(5\right)$

Wherein, and J (W, b) as defined above, KLFor openness penalty factor,Representing the average active degree (being averaged in training set) of hidden neuron j, ρ is openness Parameter, β is the weight controlling openness penalty factor.

AE neutral net adjusts parameter by constantly training makes formula (5) cost function minimum, carrys out trial learning one h_W,bThe function of ≈ x, from the data learning feature without mark, substitutes initial data feature.For marking in a large number DataIt can be finely adjusted by SAE, promotes grader Performance.When attempting solving a concrete classification problem, can be arbitrary based on these feature descriptions learning to obtain (may be less) labeled data, uses supervised learning method to complete classification.

Suppose there is size is m_lThe training set of mark(subscript l table Show " band class mark "), can be that input data find more preferable feature description.For example, it is possible to by x_l ⁽¹⁾It is input to sparse Own coding device, obtains Hidden unit activation amount a_l ⁽¹⁾.It follows that a can directly be used_l ⁽¹⁾Replace initial data x_l ⁽¹⁾(claim For " replacing representation ").Can also unite two into one, use new vector (x_l ⁽¹⁾,a_l ⁽¹⁾) replace initial data x_l ⁽¹⁾(claim For " cascade represents ").

Through conversion, training set becomes Or(depend on using a_l ⁽¹⁾Replace x_l ⁽¹⁾Or the two is merged). In practice, by a_l ⁽¹⁾With x_l ⁽¹⁾Merging generally shows more preferably.It is contemplated that internal memory and the cost of calculating, it is also possible to Use replacement operation.

Finally, a supervised learning algorithm (such as SVM, Logistic Regression etc.) can be trained, obtain Y value is predicted by one discriminant function.Prediction process is as follows: given test sample x_test, mistake before repetition Journey, is sent to sparse own coding device, obtains a_test.Then by a_test(or (x_test,a_test)) send in grader, Obtain predictive value.Concretely comprise the following steps:

1) utilize without the mark sparse own coding device of network training one.

2) a given new samples, extracts feature a by hidden layer_l ⁽ⁱ⁾。

3) training characteristics a is used_l ⁽ⁱ⁾Replace primitive character, can obtain following training dataset:

$\{(a_l^{\left(1\right)},y^{\left(1\right)}),(a_l^{\left(2\right)},y^{\left(2\right)}),...,(a_l^{\left(m_l\right)},y^{\left(m_l\right)})\}$

4) training one is from feature a^(l)To class label y⁽ⁱ⁾Logistic grader；Final grader is as shown in Figure 2.

The parameter of this model can be obtained by two training steps: at the ground floor of network, input x is mapped to hidden layer The weights W of unit activating amount a⁽¹⁾Can be obtained by sparse own coding device training process.At the second layer, Hidden unit a is reflected It is incident upon the weights W of classification output⁽²⁾Can be returned by Logistic or Softmax returns and obtains.

Summary of the invention

Goal of the invention: in order to overcome the deficiencies in the prior art, the present invention provides a kind of wide for modulus mixing amplitude modulation The anti-aliasing modem of broadcast system, provides reference for realizing analogue audio frequency broadcast to the smooth transition of digital audio broadcasting.

Technical scheme: for achieving the above object, the technical solution used in the present invention is:

Prior art is studied, it was noted that (1) demodulation of M system signal of communication is substantially simply to M Code element is classified, thus to the demodulation of MPPSK modulated signal in MPPSK/DSB-AM multiplex modulated signal, essence is The classification problem of one M system code element, principle can use DL-SAE algorithm process completely；(2) channel width Limit and the impact of intersymbol interference so that in described hybrid multiplex modulation system before and after MPPSK modulated signal code element sample it Between be provided with local correlations, contribute to SAE network and obtain the input key character of data and reconstruct input signal；(3) By selecting different signal to noise ratios and base band symbol number, modulated pattern can form enough " big data " and carry out for SAE The degree of depth learns.Therefore, train SAE network based on degree of depth learning method, allow DL-SAE learn and remember In MPPSK/DSB-AM hybrid multiplex modulation system after the inherent code element feature of MPPSK modulated signal and intersymbol interference pattern, The MPPSK modulated signal sampled value sequence of its input is carried out pattern classification and judgement, be i.e. expected to realize for The correct demodulation of MPPSK modulated signal in MPPSK/DSB-AM multiplex modulated signal.

Based on above-mentioned analysis, it is proposed that a kind of anti-aliasing modem for modulus mixing AM-broadcasting system, institute State modulus mixing AM-broadcasting system to use using MPPSK modulated signal as digital carrier to carry AM broadcast singal DSB-AM meets modulation, and the modulated signal obtained is MPPSK/DSB-AM multiplex modulated signal；This anti-aliasing modulation Demodulator includes simulated audio signal demodulator and the digital signal demodulator being in parallel；

In simulated audio signal demodulator: first the MPPSK/DSB-AM multiplex modulated signal received is believed with local oscillator Number it is multiplied；Then low pass filter or band filter is utilized to leach frequency for F_min～F_maxSimulated audio signal, real The coherent demodulation of existing audio signal；Leaching frequency followed by band filter is F_max～f_IThe component of signal of/4, i.e. divides Separate out MPPSK digital modulation baseband signal；Wherein: F_minFor the low-limit frequency of simulated audio signal, F_maxFor simulation The highest frequency of audio signal, f_IIF-FRE for receiver；

In digital signal demodulator, utilize DL-SAE neutral net that MPPSK digital modulation baseband signal is resisted Aliasing demodulates.

Preferably, the IF-FRE f of described receiver_I=465kHz, local oscillation signal is locally extracted 465kHz intermediate frequency Coherent carrier；465kHz is the wireless medium frequency standard of China's amplitude modulation broadcasting, uses this IF-FRE to make analogue audio frequency Signal still can use existing amplitude modulation broadcasting radio to listen to, it is not necessary to any change can realize simulated audio signal Non-coherent demodulation.

Concrete, described DL-SAE neutral net, is grader first with DL method by SAE neural metwork training, Make DL-SAE neutral net can produce the MPPSK of strong intersymbol interference because transmitting terminal bandwidth limits by learning and memory The wave character of modulated signal and channel effect, then directly utilize DL-SAE neutral net to MPPSK digital modulation Baseband signal carries out classification judgement, demodulates MPPSK code element.

Beneficial effect: the anti-aliasing modem for modulus mixing AM-broadcasting system that the present invention provides, relative to Prior art, has the advantage that in the following areas

1, capacity usage ratio is improved: 1. DL-SAE judgement detection fully learns and make use of intersymbol interference modulated signal The global feature of waveform and internal information, thus with simply simply utilize the threshold judgement of amplitude information and traditional matched filtering Judgement is compared, and greatly improves demodulation performance, and intersymbol interference is the most serious, compared to tradition amplitude integration judgement demodulation The advantage of device and matched filtering judgement demodulator is the biggest；2. for same chip rate, transmit and receive complex modulated letter Number can use narrower channel and receiver bandwidth, this contributes to reducing receiver noise factor, improves receiver sensitivity, And be expected to obtain higher received signal to noise ratio under same transmitting power.And receiver demodulation performance and the lifting of sensitivity All be equivalent to launch the increase of power, can be with territory, increase coverage；And if keep former communication distance index constant, then launch Power just can reduce.This is for reducing energy resource consumption and the electromagnetic pollution of broadcast system, it is achieved " green communications ", has Practical significance.

2, the availability of frequency spectrum is improved: manipulator end one-sided constriction signal spectrum bandwidth no doubt can directly improve frequency spectrum Utilization rate, but may not be feasible, because intersymbol interference may make system performance degradation to losing use value；Unless demodulation Device end, when modulated signal bandwidth constriction, remains to make demodulation performance be maintained at acceptable level.And DL-SAE demodulator Utilize non-linear, adaptability and degree of depth learning capacity that such multitiered network had, by right under stronger intersymbol interference Receive sample of signal and carry out classification judgement, it is achieved thereby that the anti-ISI solution after modulated signal of making a start " is raised speed " Adjust so that promote the availability of frequency spectrum and there is feasibility and practical significance.

3, enhance adaptation ability: 1. DL-SAE demodulator degree of depth learning capacity under " big data " support, make It can remember more signal characteristic and the characteristic of channel, thus improves the generalization ability of general nonlinearity demodulator, There is during real work higher robustness (or robustness), adapt to occasion wider；2. use classics Threshold detection or Amplitude integration is adjudicated, and is usually after taking some sampling pointwises summation near output peak value and carries out threshold judgement again, and SAE Judgement is then to the batch processing of all samplings in a code-element period, even to all sampled in n code-element period Secondary property cascading judgement, the required precision thus for sample-synchronous is far below the former, the lowest to bit synchronous requirement；③ Existing communication receiver is to compensate or eliminate the intersymbol interference that causes because band limit, first have to before demodulation to carry out channel estimation, The technical finesse such as channel equalization, liftering, and the DL-SAE demodulator of the present invention can be saved by study in advance or The step for of " merging ", it also avoid the deterioration of the front signal to noise ratio of the demodulation caused because out-of-band noise promotes；④DL-SAE Demodulator with on-line study, thus can have and adapt to modulation methods with " constant " of demodulator network topology or hardware configuration The potentiality of " ten thousand become " of formula and the characteristic of channel, beneficially demodulator and the normalization of receiver, versatility, customizable and Software radio realizes.

Accompanying drawing explanation

Fig. 1 is the example of a SAE network structure.

Fig. 2 is the example of the SAE network structure of a character identification system based on SAE, including input layer, output Layer haves three layers altogether interior；Intermediate layer is hidden layer, and for extracting the feature of input data, last layer is output layer, root Classify according to the feature extracted.

Fig. 3 is that to be that MPPSK/DSB-AM is compound adjust the hybrid multiplex modulation system theory diagram of compatible amplitude modulation broadcasting: Fig. 3 (a) Transmitter system processed；Fig. 3 (b) is the analogue system to Fig. 3 (a)；Fig. 3 (c) is that MPPSK/DSB-AM complex modulated connects Receipts machine system, it is achieved the anti-aliasing demodulation to this modulus mixing transmission signal；Fig. 3 (d) is DL-SAE classification decision device Internal theory diagram.

Fig. 4 be carrier frequency be 930kHz, MPPSK modulation parameter be K=2, M=64, r_g=0, N=K × M=128, Sample frequency f_sMPPSK digital modulation signals and the waveform correlation of multiplex modulated signal: Fig. 4 (a) of=3.6MHz are Simulated audio signal, MPPSK digital modulation signals and the time domain waveform of MPPSK/DSB-AM multiplex modulated signal； Fig. 4 (b) is the power spectrum of MPPSK digital modulation signals；Fig. 4 (c) and Fig. 4 (d) is that MPPSK/DSB-AM is multiple respectively Close time domain waveform and the power spectrum of modulated signal；Fig. 4 (e) is the MPPSK/DSB-AM after band filter molding Multiplex modulated signal launches power spectrum；Fig. 4 (f) is awgn channel signal to noise ratio when being SNR=8.6dB, and receive answers Close power spectrum and the enlarged drawing thereof of modulated signal；Frequency represents frequency.

Fig. 5 is the theory diagram of the full digital starting mode of MPPSK/EBPSK manipulator.

Fig. 6 is to MPPSK modulated signal isolated from MPPSK/DSB-AM multiplex modulated signal, DL-SAE Classification judgement demodulator (using 10,000 training datas and 10,000 test data), tradition matched filtering judgement demodulate Device and the demodulation bit error rate contrast of amplitude integration judgement demodulator.

Fig. 7 is when number of training is respectively 10,000,20,000 and 30,000 points, DL-SAE classification judgement demodulator pair The demodulation bit error rate contrast of isolated MPPSK modulated signal from MPPSK/DSB-AM multiplex modulated signal.

Fig. 8 is to MPPSK modulated signal isolated from MPPSK/DSB-AM multiplex modulated signal, DL-SAE Classification judgement demodulator (using 20,000 training datas and 20,000 test data), tradition matched filtering judgement demodulate Device and the demodulation bit error rate contrast of amplitude integration judgement demodulator.

Detailed description of the invention

Below in conjunction with the accompanying drawings the present invention is further described.

One, the modulus mixing AM-broadcasting system explanation of this case

The modulation /demodulation block diagram of the modulus mixing AM-broadcasting system of this case is as it is shown on figure 3, include modulator and demodulator two Individual part, wherein demodulator relate to DL-SAE classification decision device simultaneously；Below various piece is illustrated.

(1) manipulator of the hybrid multiplex modulation system of compatible amplitude modulation broadcasting

The hybrid multiplex modulation system of compatible amplitude modulation broadcasting uses MPPSK modulated signal to carry AM broadcast as digital carrier The DSB-AM complex modulated of signal, obtains MPPSK/DSB-AAM multiplex modulated signal expression formula as follows:

Y(t)₌A(1+k_aa(t))s_k(t) (6)

Wherein: A is the amplitude of MPPSK modulated signal, a (t) is simulated audio signal, k_aBe amplitude modulation by tone coefficient (both For preventing amplitude modulation, it is possible to control the power distribution between simulated audio signal and digital signal), s_kT () is as multiple Closing the MPPSK modulated signal of modulation carrier wave (replacing existing pure sinusoid carrier wave), it is expressed such as formula (1).According to formula (6) the MPPSK/DSB-AM complex modulator theory diagram as shown in Fig. 3 (a) is obtained.

MPPSK/DSB-AM complex modulator, is structurally modulated and digital code stream by the DSB-AM of audio signal MPPSK modulates composition, and in wherein audio signal bandwidth is limited in 4.5kHz, and system sets clocking requirement MPPSK and adjusts The character rate of signal processed is at least higher than 4.5kBd.Therefore, in order to force down MPPSK modulation after complex modulated further The signal interference to audio signal, also needs MPPSK modulated signal carrier wave both sides audio frequency before accepting modulates audio signals Power spectrum fluting in bandwidth；And after completing MPPSK/DSB-AM complex modulated, also need to carry out bandpass filtering, with full The spectral mask requirement of foot radio control agencies dictate, this is generally realized by the resonant tank of power amplifier in transmitter level.

The audio input signal of MPPSK/DSB-AM complex modulator, the brewed letter of MPPSK shown in Fig. 3 (a) and (b) Number and MPPSK/DSB-AM multiplex modulated signal waveform such as Fig. 4 (a) shown in, MPPSK modulated signal and audio frequency The power spectrum of signal is respectively as shown in Fig. 4 (b) and Fig. 4 (c), and the power spectrum of MPPSK/DSB-AM multiplex modulated signal is such as Shown in Fig. 4 (d), shown in the multiplex modulated signal power spectrum such as Fig. 4 (e) after the logical molding filtration of band of making a start.In view of actual measurement It is resolution to be adjusted to and audiofrequency spectrometer resolution shown in the restriction of time-frequency spectrometer resolution, Fig. 4 (b), Fig. 4 (d) and Fig. 4 (e) Power spectrum chart when consistent 1～5Hz.And the theoretical power (horse-power) spectrum sideband under environment described in this case, will be than institute in Fig. 4 Show is lower.The PSD sideband of the digital modulation signals shown in Fig. 4 (b) at the 9kHz of simulation main signal (with carrier frequency is Center ± 4.5kHz) in bandwidth less than carrier wave 50dB, therefore analogue audio frequency can be realized and the same of digital signal is broadcast.

(2) demodulator of the hybrid multiplex modulation system of compatible amplitude modulation broadcasting

The most complete backward compatible existing commodity AM of MPPSK/DSB-AM complex modulated receiver that this case is proposed receives Sound machine, shown in theory diagram such as Fig. 3 (c) of whole receiver (i.e. radio) system.Specific works process is as follows:

(1) wireless antenna tuning loop is from f_rThe medium wave amplitude modulation frequency range of=531kHz～1602kHz selects institute The MPPSK/DSB-AM multiplex modulated signal (shown in power spectrum such as Fig. 4 (f)) of broadcast listening channel, through preamplifier After amplification, being mixed (being multiplied and bandpass filtering) with the local oscillation signal from local oscillator, obtaining frequency is f_I=465kHz Intermediate frequency MPPSK/DSB-AM signal, through intermediate frequency amplify after be divided into two-way to export.Envelope detector is given on one tunnel, Can demodulate simulated audio signal and deliver to earphone or speaker sounding, this is entirely directly continues to use existing wireless function, The sound quality that demodulation recovers is also suitable with during employing pure sinusoid carrier signal modulation, embodies backward compatible existing simulation The feasibility of AM broadcasting system.

(2) another road intermediate frequency MPPSK/DSB-AM multiplex modulated signal after amplifying is used for demodulating MPPSK number Word modulation intelligence, this partial function is that Digital AM Broadcasting receiver in the future must add.Multiple for this road intermediate frequency Closing modulated signal, transformed to base band first with coherent demodulation technology, the core devices of coherent demodulator is in one Frequently analog multiplier, it is achieved two input signals being multiplied: is exactly MPPSK/DSB-AM intermediate-freuqncy signal, another Individual, extract from MPPSK/DSB-AM intermediate-freuqncy signal, the purest and with in MPPSK/DSB-AM Frequently signal is with the coherent carrier of frequency homophase.Two paths of signals is multiplied in multiplier and filters 2 harmonics, and it is right i.e. to achieve The coherent demodulation of MPPSK/DSB-AM multiplex modulated signal, it will be transformed into base band by intermediate frequency.

(3) coherent demodulator (at this for multiplier) is output as MPPSK digital modulation signals and simulated audio signal Mixing (superposition), the two is complete aliasing in time waveform, is then less than due to audio signal frequency spectrum on frequency spectrum 4.5kHz, and aforementioned " manipulator of the hybrid multiplex modulation system of compatible amplitude modulation broadcasting " design requires MPPSK modulation letter Number character rate 4.5kBd to be exceeded, which ensure that the fundamental frequency of MPPSK baseband signal higher than 4.5kHz, Thus can utilize corresponding wave filter separation simulation audio signal and digital baseband signal, utilize low pass the most exactly Wave filter audio signal is directly taken out (currently in order to backward compatible existing AM radio can, and number in the future Word broadcast receiver then can save based on this in Fig. 3 (c) for realizing the envelope detector of non-coherent demodulation), with Shi Liyong high pass (or band logical) filters filter goes out MPPSK modulated signal (or the main frequency of MPPSK modulated signal Rate composition).

(4) isolated base band MPPSK modulated signal is sent into DL-SAE classification decision device, therefrom demodulate former The modulation data code flow begun.

(3) DL-SAE classification decision device

This case is intended in the case of stronger intersymbol interference, the difference being degrading in the overall waveform feature of M system code element, It is supplied to SAE learn further, train and differentiate.Therefore, SAE is adjudicated detection method to be applied to The demodulation of MPPSK modulated signal in MPPSK/DSB-AM hybrid system, is the key technology content of the present invention.

But, SAE itself is not specialized in this case, but for being successfully used for image recognition and hand writing system divides The SAE network structure in the fields such as class is appropriately modified, and is allowed to MPPSK modulation be applicable to training hybrid multiplex modulation system The sample sequence of signal, proposes MPPSK modulated signal anti-ISI demodulator based on DL-SAE, theory diagram As shown in Fig. 3 (d).Here SAE is trained for grader, the whole code element received is identified.From from not Carry out degree of depth study with signal to noise ratio in " big data " sample that communicates under intersymbol interference environment, extract and remember with code Between the wave character of modulated signal filter response of interference and internal association.SAE exports one for each input symbols Indicating the numeral of its classification, as represented " 0 " code element with 0, representing " 1 " code element with 1, until representing " M code with M Unit ", then go to control to export corresponding local standard symbol waveform by this numeral, training can be completed.After training successfully, To newly inputted signal sample sequence, classification judgement is carried out with regard to the available network that degree of depth has succeeded in school, demodulates MPPSK Data symbols.

Visible, DL-SAE demodulator principle proposed by the invention can be used for demodulating various digital modulation signals.

Two, a specific embodiment based on this case thought

(1) selection of MPPSK modulation parameter

According in this specification background technology about described in the content of " MPPSK transmission system ", selecting MPPSK Formula (2) should be met during modulation parameter as far as possible, i.e. take r_g=0 and N=M × K, so it is completely eliminated MPPSK Adjust the discrete line spectrum in power spectrum signal, for simulated audio signal and other broadcasting station of adjacent channel of same channel The interference of signal, can be lower.

Transmitting carrier frequency in view of China's medium wave amplitude modulation frequency range is defined as 531kHz～1602kHz, therefore for describing the problem, The present embodiment takes Beijing platform f in its frequency range stage casing_c=930kHz, and take K=2, M=64, N=128, now The symbol rate of MPPSK modulation is f_c/ N=930/128=7.265625kBd, aforementioned must exceed audio signal highest frequency The requirement of 4.5kHz is met, and transmission code rate is then up to R_bMPPSK=(f_c log₂M)/N=43.59375kbps.Cause This, is even if taking out half code check in actual applications carries out chnnel coding in the future, to be further ensured that the reliability of system, We still can obtain the clean code check of about 21.8kbps for transmitting data, or carries out the frequency modulation tonequality numeral of 16kbps code check Sound radio.

(2) realization of MPPSK manipulator

In order to easy to adjust, the present embodiment uses Digital Way as shown in Figure 5 to realize for MPPSK manipulator: The MPPSK sequence of symhols with M kind value is utilized to control MUX, from M group MPPSK modulation waveform numeral Selecting corresponding with current symbol in sample, sending digital to analog converter (DAC) to be directly changed into carrier frequency is f_cSimulation MPPSK modulated signal exports.

(3) the DSB-AM modulation of audio signal

Owing to described MPPSK modulated signal is analogous to the constant envelope signal of sine wave, therefore, for simulation during emulation The realization of audio signal MPPSK/DSB-AM complex modulated, after only need to adding DC component in audio signal and MPPSK modulated signal is multiplied and retains whole frequency component, as shown in Fig. 3 (b).Add DC component be in order to Avoid the amplitude modulation of crossing of audio signal during emulation, and the DC component added is the biggest, in MPPSK/DSB-AM modulated signal Carrier component the strongest, and this DC component is exactly the direct current biasing electricity of its modulating stage in actual AM broadcast transmitter Flat, and amplitude-modulation index can be by changing amplitude modulation factor k in formula (6)_aRegulate.The present embodiment controls audio frequency letter Number absolute peak is added DC component amplitude 90%, i.e. k in formula (6)_a=0.9.The complex modulated letter of multiplier output Number amplify and after filtering through broadcast transmitter power amplifier, can deliver to launching tower (antenna) on emission, as schemed (as long as and in emulating, exporting the digital sample values of MPPSK/DSB-AM modulated signal) shown in 3 (b).

(3) realization of MPPSK demodulator

Shown in theory diagram such as Fig. 3 (c) of whole receiver (i.e. radio) system, including the analogue audio frequency letter being in parallel Number demodulator and digital signal demodulator.In simulated audio signal demodulator: the MPPSK/DSB-AM received is multiple Close modulated signal to be first multiplied with local oscillation signal；Then utilize low pass filter or band filter to leach frequency to be F_min～F_maxSimulated audio signal, it is achieved the coherent demodulation of audio signal；Frequency is leached followed by band filter For F_max～f_IThe component of signal of/4, i.e. isolates MPPSK digital modulation baseband signal.In digital signal demodulator, Utilize DL-SAE neutral net that MPPSK digital modulation baseband signal is carried out anti-aliasing demodulation.

Shown in theory diagram such as Fig. 3 (d) of DL-SAE classification decision device, in order to utilize SAE to newly inputted MPPSK Modulated signal sample sequence carries out Classification and Identification, first has to be trained its interconnective weights coefficient of intrinsic nerve unit, SAE to be allowed learns and remembers object to be classified or pattern.Being trained for without supervised training mould of SAE network Type, it is achieved output, equal to input, is i.e. analogous to Training model using input as label and is trained, the present embodiment Directly DL-SAE is carried out Matlab training and emulation.

(1) rule of DL-SAE demodulator design

Owing to input signal during DL-SAE demodulator real work is containing interchannel noise, thus also need when training Very important person is for adding certain noise or disturbance, and one is the real work situation in order to meet demodulator in the future, and two is that it can be straight Connect and affect the generalization ability that SAE demodulator is final.In this case, training noise size according to intersymbol interference intensity and Fixed, but there is no theory at present and can follow, the present embodiment can only obtain some Experience norms and specific practice according to substantial amounts of emulation:

1. channel circumstance is the most severe, and intersymbol interference is the most serious, and training noise should be the least.This point is it can be appreciated that because right In any communication system, being the most all that reception signal is the most severe with reception environment, receptivity is the poorest (even cannot work Make), thus the noise that extra interpolation is bigger now also it is no need for again when SAE trains；

2. the method that can take many experiments, the network the most best to obtain generalization ability.

(2) selection of DL-SAE demodulator simulation parameter

Modulation still uses the defined MPPSK modulation improved of formula (1), carrier frequency f_c=930kHz, and take K=2, M=64, N=K × M=128, r_g=0, sample frequency 3.6MHz.

(3) determination of DL-SAE demodulator number of training

DL-SAE algorithm is a class " big data " algorithm, and its effectiveness is directly related with amount of training data, in theory, Amount of training data is the biggest, and the accuracy of DL-SAE classification is the highest.But huge data volume will necessarily reduce computational efficiency. This case compared for 10,000 training datas, 10,000 test data and 20,000 training datas, the solutions of 20,000 test data Adjust the bit error rate, consider the factors such as computational efficiency, use 20,000 training datas, 20,000 test data to carry out reality Test.

(4) select SAE structure and iterations and carry out SAE network training

1. SAE structure relates generally to the number of plies of network and the unit number of every layer, and the hidden layer of usual network is the most, every layer of institute The unit number contained is the most, and the training effect of network is the best, but the too much network number of plies and unit number there will be over-fitting. Through test of many times certification, for the data of the present embodiment, the SAE network the most preferably used is 4 Rotating fields: 512-400-200-64, including the input layer containing 512 unit, 1,200 unit of hidden layer of 400 unit hidden Layer 2, and the output layer of 64 unit, i.e. the classification output of M=64 system.For obtaining best classifying quality, warp Test of many times, taking relevant network parameter is: add coefficient of making an uproar be 0.1, dropout parameter be 0.1, weight attenuation quotient It is 1 × 10^-4。

2. increase iterations and can improve the learning capacity of SAE, but will take considerable time, for balance quality and training Cost, the present embodiment have employed 40 iteration and is trained SAE.

" big data " characteristic learnt by the degree of depth, for different data, will have different optimum DL-SAE models, Therefore the DL-SAE model that the present embodiment uses is merely to illustrate the present embodiment, is not limited to the present invention.

(5) performance simulation

DL-SAE classification judgement and tradition matched filtering judgement when Fig. 6 is 10,000 training datas and 10,000 test data And the bit error rate contrast of the MPPSK demodulation output of amplitude integration judgement.As seen from Figure 6, DL-SAE demodulator Performance, than traditional matched filtering judgement demodulator and amplitude integration judgement demodulator, at least improves 1 order of magnitude.

Fig. 7 is number of training when being respectively 10,000,20,000 and 30,000 points, DL-SAE classification judgement demodulator to from The demodulation bit error rate contrast of isolated MPPSK modulated signal in MPPSK/DSB-AM multiplex modulated signal.By Fig. 7 It can be seen that sample number be bit error rate when 20,000 and 30,000 will less than 10,000 time the bit error rate, but be more or less the same, And when signal to noise ratio is close to 25dB, the bit error rate of 20,000 samples and 30,000 samples is essentially identical.Visible, work as noise Time the highest, the demodulation performance that increase sample number brings promotes and inconspicuous, then is integrated into the factors such as computational efficiency, this reality Executing example finally to choose at 20,000 and carry out emulation experiment, the most final demodulation performance contrast is shown in Fig. 8.

As seen from Figure 8, the strong intersymbol interference caused due to transmitting terminal molding filtration, even if result at high s/n ratio Under, the bit error rate of the judgement of matched filtering device and the judgement of amplitude integration does not all reach 10^-3Magnitude, then promote signal to noise ratio and also cannot Improve；The decision algorithm but the DL-SAE that this specification is invented classifies, but can go out preferably from mass data learning Classifying face, makes correct classification to the MPPSK symbol waveform of multi-system, demodulation bit error rate is down to 10^-4Magnitude.

The above is only the preferred embodiment of the present invention, it should be pointed out that: for those skilled in the art For, under the premise without departing from the principles of the invention, it is also possible to make some improvements and modifications, these improvements and modifications are also Should be regarded as protection scope of the present invention.

Claims (3)

1. for an anti-aliasing modem for modulus mixing AM-broadcasting system, described modulus mixing amplitude modulation broadcasting System uses the DSB-AM carrying AM broadcast singal as digital carrier using MPPSK modulated signal to meet modulation, The modulated signal obtained is MPPSK/DSB-AM multiplex modulated signal；It is characterized in that: this anti-aliasing modem Including the simulated audio signal demodulator being in parallel and digital signal demodulator；

In simulated audio signal demodulator: first the MPPSK/DSB-AM multiplex modulated signal received is believed with local oscillator Number it is multiplied；Then low pass filter or band filter is utilized to leach frequency for F_min～F_{m ax}Simulated audio signal, real The coherent demodulation of existing audio signal；Leaching frequency followed by band filter is F_max～f_IThe component of signal of/4, i.e. divides Separate out MPPSK digital modulation baseband signal；Wherein: F_minFor the low-limit frequency of simulated audio signal, F_maxFor simulation The highest frequency of audio signal, f_IIF-FRE for receiver；

In digital signal demodulator, utilize DL-SAE neutral net that MPPSK digital modulation baseband signal is resisted Aliasing demodulates.

Anti-aliasing modem for modulus mixing AM-broadcasting system the most according to claim 1, it is special Levy and be: the IF-FRE f of described receiver_I=465kHz, local oscillation signal is that locally extracted 465kHz intermediate frequency is concerned with Carrier wave.

Anti-aliasing modem for modulus mixing AM-broadcasting system the most according to claim 1, it is special Levy and be: described DL-SAE neutral net, be grader first with DL method by SAE neural metwork training, make Obtain DL-SAE neutral net and can produce the MPPSK tune of strong intersymbol interference because the restriction of transmitting terminal bandwidth by learning and memory The wave character of signal processed and channel effect, then directly utilize DL-SAE neutral net to MPPSK digital modulation base Band signal carries out classification judgement, demodulates MPPSK code element.