Accumulate 8 core DSP on piece Parallel Implementation methods of cross-correlation envelope alignment
Technical field
The present invention relates to multi-core DSP parallel methods, more particularly to a kind of intensive of multi-core DSP Parallel Implementation, one by one arteries and veins
The method for rushing the accumulation cross-correlation envelope alignment of recursive operation.
Background technology
In inverse synthetic aperture radar imaging (ISAR), the good and bad of envelope alignment can be great to the generation of last image quality
It influences.For adjacent twice return, the corner of target is generally less than 0.01o, therefore their real envelope is quite similar, related
Coefficient is typically up to more than 0.95.Based on the above-mentioned fact, domestic and international many experts and scholars have carried out envelope alignment algorithm largely
Research, can be seen that mainly from the document published has following three kinds of approach:
1st, cross-correlation function maximal criterion;
2nd, Minimum Entropy criteria;
3rd, the least euclidean distance criteria.
The invention mainly relates to it is most studied, apply to obtain widest the first criterion, and employ and a kind of can have
Effect avoids the accumulation reference signal cross-correlation envelope alignment computational methods of envelope drift and kick error.Computational methods maximum
Feature is that its reference signal is not some single echo-signal, but weights accumulation real-time update.It in this way can be largely
Ground melts weight of the abnormal signal in reference signal, can effectively eliminate shadow of the target echo mesorelief part to motion compensation
It rings;Signal more neighbouring with echo to be aligned in reference signal accounts for higher weights simultaneously, can obtain better alignment effect.It lacks
Point is exactly that this algorithm can only be by pulse echo sequentially operation, it is impossible to which parallel computation improves efficiency.
In recent years, requirement of the various algorithms to DSP platform increasingly improves, and TI companies are proposed industry in November, 2010
The highest multi-core DSP processing chip TMS320C6678 of energy, has surmounted the every other DSP core of industry, has become first acquisition at one stroke
The DSP that highest fixed-point performance and floating-point performance are divided equally.It possesses 8 C66x kernels using Keystone frameworks, each kernel
Highest is operated on the highest frequency of 1.25GHz, provides the high-performance of 160GFLOPs.Have benefited from the powerful performances of C6678, it will
It is applied to the envelope alignment in ISAR, can effectively solve the problems, such as its big data quantity, complicated algorithm.
Invention content
The technical problems to be solved by the invention are just to provide a kind of 8 core DSP on pieces for accumulating cross-correlation envelope alignment simultaneously
Row implementation method is to realize under the premise of algorithm performance is not influenced, by the accumulation reference signal cross-correlation envelope pair of sequence execution
Neat algorithm is concurrently realized on 8 core DSP, greatly improves its operation time.
The present invention solves the technical problem, the technical solution adopted is that, 8 core DSP pieces of accumulation cross-correlation envelope alignment
Upper Parallel Implementation method, includes the following steps:
Step 1, DSP core 0 make accumulation cross-correlation envelope pair to the pulse echo of the preceding Z PRF of a frame radar return data
It is neat to handle to obtain stable global reference signal refg, it is to be shared in the follow-up alignment algorithms of core 0-7, wherein Z is 8 multiple;
Step 2, DSP core 0-7 are according to global reference signal refgThe parallel accumulation cross-correlation for handling follow-up 8 PRF respectively
Envelope alignment, each core respectively obtain the echo-signal after alignment and local reference signal refI, I=[0,1,2,3,4,5,6,
7], represent when pronucleus serial number, enter step 3;
Step 3, DSP core 0 detect whether all pulse signals of a frame have all been aligned, and are the echo letters exported after alignment
Number, otherwise enter step 4;
Step 4, eight reference signal ref that DSP core 0 is exported using eight cores【0-7】The global reference signal of update, repeats step
2。
Specifically, in the step 1, DSP core 0 accumulates the pulse echo of the preceding Z PRF of a frame radar return data
Cross-correlation envelope alignment handles to obtain stable global reference signal refg, include the following steps:
For the radar return data that a pending frame N*M is tieed up, wherein, N is echo impulse number, and M is each pulse
Contained range cell number;Using first pulse as initial reference signal ref, second pulse echo is taken to be used as and works as pre-echo
It is denoted as st;
A1,0 calculating benchmark signal ref of core are with working as pre-echo stCross-correlation function R, and obtain cross-correlation function peak value
RMAXAnd its corresponding coordinate is denoted as XMAX, enter step B1;
B1, core 0 take out peak point (X respectivelyMAX,RMAX) the adjacent cross-correlation function value in left and right and its corresponding coordinate difference
It is denoted as (X1,R1) and (X2,R2), this 3 points can fit the downward quadratic term curve of a recess, calculate the maximum of the curve
The corresponding coordinate of value pointEnter step C1;
C1, core 0 utilize coordinate X' construction compensation terms at this time:And with as pre-echo stDot product (respective distances
Unit is multiplied)With the current PRF echo s after being alignedt' and store, enter step D1;
D1, according to the current PRF echo s after alignmentt' calculate new reference signal:Ref=γ ref+st', wherein,
Update weight coefficient 0≤γ≤1;So far the envelope alignment of current PRF echo is completed, and enters step E1;
E1, core 0 take out next pulse echo, according to new reference signal, step A are repeated, until preceding Z echo data
Alignment is fully completed, obtains stable global reference signal refg。
Further, it in the parallel processing of accumulation cross-correlation envelope alignment algorithm, further includes and is become using fast Fourier
(FFT) is changed to reference signal ref with working as pre-echo stCarry out interpolation processing:
It is assumed that the required precision of envelope alignment is 1/8 range cell, then needs to reference signal ref and work as pre-echo
st8 times of interpolation are carried out, the FFT of different points twice can be used in dsp to realize 8 times of interpolation, i.e.,:IFFT[FFT(st,M),
8M], wherein, M is the range cell number contained by each pulse.
Further, core 0 improves cross-correlation function computational efficiency using Fast Fourier Transform (FFT) (FFT);
Calculating benchmark signal ref is with working as pre-echo stCross-correlation function R when, can be realized using 3 FFT, i.e.,:R=
IFFT[FFT(st).*FFT(ref)]。
Specifically, in the step 2-4, include the following steps:
A2, core 0-7 calculate global reference signal ref respectivelygWith as pre-echo stCross-correlation function R, and obtain cross-correlation
Peak of function RMAXAnd its corresponding coordinate is denoted as XMAX, enter step B2;
B2, core 0-7 take out peak point (X respectivelyMAX,RMAX) the adjacent cross-correlation function value in left and right and its corresponding coordinate point
(X is not denoted as it1,R1) and (X2,R2), this 3 points can fit the downward quadratic term curve of a recess, calculate the curve most
The corresponding coordinate of big value pointEnter step C2;
C2, core 0-7 are utilized respectively coordinate X' construction compensation terms at this time:And with as pre-echo stDot product is (right
Range cell is answered to be multiplied)With the current PRF echo s after being alignedt' and store, with this eight
Core respectively obtains the pulse echo after eight alignment, and respectively according to the current PRF echo s after alignmentt' calculate corresponding core
Partial reference signal:refI=st';So far the envelope alignment of current PRF echo is completed, and enters step D2;
D2, the partial reference signal update overall situation reference signal that core 0 is exported according to eight cores:
refg=γ8refg+γ7ref7+γ6ref6+γ5ref5+γ4ref4+γ3ref3+γ2ref2+γ1ref1+
ref0;0≤γ≤1;
Wherein, the partial reference signal that ref0 is calculated for core 0, the partial reference signal that ref1 is calculated for core 1, ref2
For the partial reference signal that core 2 calculates, ref3 is the partial reference signal that core 3 calculates, and ref4 is the part that core 4 calculates
Reference signal, ref5 are the partial reference signal that core 5 calculates, and ref6 is the partial reference signal that core 6 calculates, and ref7 is core
The 7 partial reference signals calculated;
Next pulse echo is taken out, according to new global reference signal, step A2 is repeated, until a frame echo data is complete
Complete alignment, the echo-signal after output alignment in portion.
Further, it in the parallel processing of accumulation cross-correlation envelope alignment algorithm, further includes and utilizes Fourier's variation pair
Global reference signal refgWith as pre-echo stCarry out interpolation processing:
It is assumed that the required precision of envelope alignment is 1/8 range cell, then need to global reference signal refgWith it is current
Echo st8 times of interpolation are carried out, the FFT of different points twice can be used in dsp to realize 8 times of interpolation, i.e.,:IFFT[FFT(st,
M), 8M], wherein, M is the range cell number contained by each pulse.
Further, core 0 improves cross-correlation function computational efficiency using FFT;
Calculate global reference signal refgWith as pre-echo stCross-correlation function R when, can be realized using 3 FFT, i.e.,:
R=IFFT [FFT (st).*FFT(refg)]。
Specifically, the core 0 of DSP is used as main core, transmitting-receiving radar return data, control core 0-7 operations and realization accumulation are mutual
Envelope alignment is closed to calculate.
Further, the communication between core 0-7 and collaborative work are by changing one of the shared memory for being located at DSP
For Status Flag variable flag [8] come what is realized, which is 8 dimension groups of int types, corresponds to the Status Flag of each core respectively,
And each core can access and change the variable, initial value is all 0.
Further, after DSP core 0 obtains stable global reference signal, core 0 changes the Status Flag of all cores, makes it
It is incremented by 1;When each pyrene detects that the Status Flag of this core is incremented by 1, then after being handled respectively parallel using global reference signal
Continue the accumulation cross-correlation envelope alignment of 8 PRF, and respectively obtain echo and 8 partial reference signals after 8 alignment simultaneously,
Then the Status Flag of this core is made to be incremented by 1;When core 0 detects that the Status Flag of 8 cores has all increased by 1, then 8 local bases are utilized
The global reference signal of calibration signal update, the Status Flag of all cores is incremented by 1 by core 0 again after having updated, and repeats step 2.
The beneficial effects of the invention are as follows:The present invention makes full use of the high performance characteristics of 8 core DSP, will build up on cross-correlation envelope
Alignment algorithm is concurrently realized on 8 core DSP, is reached and is improved algorithm computational efficiency purpose;Meanwhile the present invention is to be aligned by a frame
The preceding Z pulse echo elder generation sequential operations of radar return data obtain metastable reference signal, can guarantee whole frame in this way
The alignment effect of echo;On the other hand, reference signal is not some single echo-signal, but weights accumulation real-time update
, weight of the abnormal signal in reference signal can be largely melted in this way, can effectively eliminate target echo mesorelief portion
Divide the influence to motion compensation;Signal more neighbouring with echo to be aligned in reference signal accounts for higher weights simultaneously, can obtain
Better alignment effect.
Description of the drawings
Fig. 1 show 8 cores in 8 core DSP on piece Parallel Implementation embodiments of the method for this hair name accumulation cross-correlation envelope alignment
DSP (TMS320C6678) on piece Parallel Implementation method realizes block diagram;
Fig. 2 show DSP in 8 core DSP on piece Parallel Implementation embodiments of the method for this hair name accumulation cross-correlation envelope alignment
Upper allocation process diagram;
Fig. 3 is shown in 8 core DSP on piece Parallel Implementation embodiments of the method for this hair name accumulation cross-correlation envelope alignment mutually
It closes function maxima binomial interpolation and chooses maximum of points schematic diagram;
Fig. 4 is first arteries and veins in 8 core DSP on piece Parallel Implementation embodiments of the method for this hair name accumulation cross-correlation envelope alignment
Refunds wave process chart;
Fig. 5 is shown preceding 16 in 8 core DSP on piece Parallel Implementation embodiments of the method for this hair name accumulation cross-correlation envelope alignment
A pulse process chart.
Specific embodiment with reference to embodiments is described in further detail the above of the present invention again.
But the range that this should not be interpreted as to the above-mentioned theme of the present invention is only limitted to following example.Think not departing from the above-mentioned technology of the present invention
In the case of thinking, the various replacements or change made according to ordinary skill knowledge and customary means should all be included in this hair
In bright range.
Specific embodiment
Detailed description of the present invention technical solution with reference to the accompanying drawings and embodiments:
The present invention is for the accumulation reference signal cross-correlation that can effectively avoid envelope drift and kick error in the prior art
The computational methods of envelope alignment can only be by pulse echo sequentially operation, it is impossible to which parallel computation improves the problem of efficiency, provides one kind
8 core DSP on piece Parallel Implementation methods of cross-correlation envelope alignment are accumulated, first, DSP core 0 is to the preceding Z of a frame radar return data
The pulse echo of a PRF makees the processing of accumulation cross-correlation envelope alignment to obtain stable global reference signal refg, after being core 0-7
It is shared in continuous alignment algorithm, wherein Z is 8 multiple;Secondly, DSP core 0-7 is according to global reference signal refgIt handles respectively parallel
The accumulation cross-correlation envelope alignment of follow-up 8 PRF, each core respectively obtain the echo-signal after alignment and local reference signal
refI, I=[0,1,2,3,4,5,6,7] represented when pronucleus serial number, and then whether all DSP core 0 detects all pulse signals of a frame
It has been be aligned that, be, exported the echo-signal after alignment, eight reference signal ref that otherwise SP cores 0 are exported using eight cores【0-7】More
New overall situation reference signal, repeats the above steps.The present invention makes full use of the high performance characteristics of 8 core DSP, will build up on cross-correlation envelope
Alignment algorithm is concurrently realized on 8 core DSP, is reached and is improved algorithm computational efficiency purpose;Meanwhile the present invention is to be aligned by a frame
The preceding Z pulse echo elder generation sequential operations of radar return data obtain metastable reference signal, can guarantee whole frame in this way
The alignment effect of echo;On the other hand, reference signal is not some single echo-signal, but weights accumulation real-time update
, weight of the abnormal signal in reference signal can be largely melted in this way, can effectively eliminate target echo mesorelief portion
Divide the influence to motion compensation;Signal more neighbouring with echo to be aligned in reference signal accounts for higher weights simultaneously, can obtain
Better alignment effect.In addition, interpolation processing and calculating cross-correlation function present invention profit in cross-correlation envelope alignment is accumulated
Computational efficiency is improved with Fast Fourier Transform (FFT) (FFT).
Embodiment
This example is proposed the highest multi-core DSP processing chip of industry performance in November, 2010 with TI companies
For TMS320C6678, the invention content of the present invention is specifically described, describes present disclosure for convenience, made first following
Technical term defines:
Define 1:Inverse Synthetic Aperture Radar (ISAR) is imaged;
ISAR is to improve lateral resolution by the relative motion (be usually radar motionless) between radar and target, using big
Bandwidth improves distance resolution, and two-dimensional imaging is carried out to general radar target, such as aircraft, naval vessel, guided missile with this.
Define 2:One frame data;
One frame data refer into several pulses of the width ISAR as needed for.
Define 3:Envelope alignment;
Target to be imaged relative to radar there are translation component and rotative component, for target relative to radar penetrate by translation component
The posture of line remains unchanged, and makees translational motion.Translation component can make target echo envelope shifting, this is to being finally imaged no tribute
It offers, needs to fall the compensation of its effect.Translational compensation includes envelope alignment again and first phase corrects, and envelope alignment is exactly by a frame number
According to the target echo reality envelopes of all pulses be aligned one by one.
Define 4:Shared memory;
Shared memory in the present invention is C6678 multinuclears shared storage area, refers to that address is arrived for 0x0C000000
The storage region of the common 4M of 0x0C3FFFFF accesses for 8 cores.
This example under the premise of algorithm performance is not influenced, calculate by the accumulation reference signal cross-correlation envelope alignment that sequence is performed
Method is concurrently realized on C6678, greatly improves its operation time, as shown in Figure 1, C6678 cores 0 are first to a frame radar return number
Preceding Z before, wherein Z is 8 multiple;This example makees accumulation cross-correlation envelope for the pulse echo of 16 (0-15) PRF
For registration process to obtain metastable reference signal, then core 0 changes the Status Flag of all cores, increments it by 1;Second
Step, the core 0-7 of DSP once detect that this core flag bit has been changed to desired value, then parallel to handle follow-up 8 respectively (as worked as
During flag [0-8]=1, core 0-7 handles the 16-23 pulse echo respectively) the accumulation cross-correlation envelope alignment of PRF, it has been calculated
Echo-signals and reference signal of the Cheng Houhui after the program alignment of setting address output, while corresponding flag bit is incremented by 1 again;
Third walks, and core 0 detects the flag bit state of all cores, once all having increased by 1, then 8 cores are complete this operation, utilize at this time
The reference signal of this 8 core outputs calculates new reference signal;Cycle execution second and third step are until all arteries and veins of this frame
Signal is rushed all to be aligned.Finally, the Status Flag of all cores is all set to 0 by core 0, i.e. flag [0-7]=0 can perform entire DSP
The envelope alignment of next frame data.Wherein, core 0 performs transceiving data, control as shown in Fig. 2, using the core 0 of multi-core DSP as main core
The operation of other cores and calculating is made, and the communication and collaborative work between multinuclear are that the shared storage of DSP is located at by changing
For one Status Flag variable flag [8] in area come what is realized, which is 8 dimension groups of int types, corresponds to the shape of each core respectively
State mark, and each core can access and change the variable, initial value is all 0.The high performance characteristics of 8 core DSP are made full use of, it will
Accumulation cross-correlation envelope alignment algorithm is concurrently realized on 8 core DSP, is reached and is improved algorithm computational efficiency purpose;Meanwhile this hair
Preceding 16 pulse echo elder generations sequential operation of the bright radar return data that a frame is to be aligned obtains metastable reference signal,
It can guarantee the alignment effect of whole frame echo in this way;On the other hand, reference signal is not some single echo-signal, but is added
Power accumulation real-time update, weight of the abnormal signal in reference signal can be largely melted in this way, can effectively eliminate mesh
Mark influence of the echo mesorelief part to motion compensation;Signal more neighbouring with echo to be aligned in reference signal accounts for higher simultaneously
Weights, better alignment effect can be obtained.
Lower mask body illustrates the present invention using 256 pulse echos of a frame as echo data to be aligned;
First, initiation parameter and shared memory variable is opened up.Initiation parameter has:One frame pending data has 256
A pulse echo, each pulse echo include 1024 sampled points.Status Flag bit array is set in C6678 shared memories,
And assign initial value ' 0 ', i.e. flag [0-7]=0.The identical reference signal space of 9 length is opened up simultaneously, is envelope pair respectively
The reference signal for using to obtain after reference signal and each operations of core 0-7 in neat.In addition, treat the data and envelope of envelope alignment
Data after alignment are all placed on the external memory space of C6678.
Secondly, for the envelope alignment of preceding 16 pulse echos of a frame pending data, core 0 performs;Core 0 receives one
Frame echo data to be aligned, and using the receive the 0th pulse echo as the first reference signal, opened from the 1st pulse echo
Beginning sequence performs accumulation cross-correlation envelope alignment algorithm, until preceding 16 pulses have all been aligned, then by all core mark positions
‘1’.Wherein, specifically the processing of accumulation cross-correlation envelope alignment specifically includes:
For the radar return data that a pending frame N*M is tieed up, wherein, N is echo impulse number, and M is each pulse
Contained range cell number;Using first pulse as initial reference signal ref, second pulse echo is taken to be used as and works as pre-echo
It is denoted as st, it is specific as shown in Figure 3;
A1,0 calculating benchmark signal ref of core are with working as pre-echo stCross-correlation function R, and obtain cross-correlation function peak value
RMAXAnd its corresponding coordinate is denoted as XMAX, enter step B1;
B1, core 0 take out peak point (X respectivelyMAX,RMAX) the adjacent cross-correlation function value in left and right and its corresponding coordinate difference
It is denoted as (X1,R1) and (X2,R2), this 3 points can fit the downward quadratic term curve of a recess, calculate the maximum of the curve
The corresponding coordinate of value pointEnter step C1;
C1, core 0 utilize coordinate X' construction compensation terms at this time:And with as pre-echo stDot product (respective distances
Unit is multiplied)With the current PRF echo s after being alignedt' and store, enter step D1;
D1, according to the current PRF echo s after alignmentt' calculate new reference signal:Ref=γ ref+st', wherein,
Update weight coefficient 0≤γ≤1;So far the envelope alignment of current PRF echo is completed, and enters step E1;
E1, core 0 take out next pulse echo, according to new reference signal, step A are repeated, until preceding Z echo data
Alignment is fully completed, obtains stable global reference signal refg。
As shown in Figures 4 and 5, it in the parallel processing of above-mentioned accumulation cross-correlation envelope alignment algorithm, further includes using fast
Fast Fourier transformation (FFT) is to reference signal ref with working as pre-echo stCarry out interpolation processing:
It is assumed that the required precision of envelope alignment is 1/8 range cell, then needs to reference signal ref and work as pre-echo
st8 times of interpolation are carried out, the FFT of different points twice can be used in dsp to realize 8 times of interpolation, i.e.,:IFFT[FFT(st,M),
8M], wherein, M is the range cell number contained by each pulse.Meanwhile core 0 is improved mutually using Fast Fourier Transform (FFT) (FFT)
Close function computational efficiency;
Calculating benchmark signal ref is with working as pre-echo stCross-correlation function R when, can be realized using 3 FFT, i.e.,:R=
IFFT[FFT(st).*FFT(ref)]。
Then, loop body is established.In loop body, it is that (cycling condition is n to set up cycle-index:For (n=0;n<30;n+
+)), core 0- cores 7 are performed simultaneously:When each core is detected (as pronucleus serial number CoreNum) when the state flag bit of pronucleus is
During flag [CoreNum]=2n+1, just take out corresponding (16+8n+CoreNum) a pulse echo and do accumulation cross-correlation envelope
Then alignment provides corresponding alignment back echo and when pronucleus reference signal, while updates its state flag bit:flag
[CoreNum]=flag [CoreNum]+1;Wherein, the meter of the processing of cross-correlation envelope alignment and preceding 16 pulses is specifically accumulated
Calculation mode is consistent, specifically includes following steps:
A2, core 0-7 calculate global reference signal ref respectivelygWith as pre-echo stCross-correlation function R, and obtain cross-correlation
Peak of function RMAXAnd its corresponding coordinate is denoted as XMAX, enter step B2;
B2, core 0-7 take out peak point (X respectivelyMAX,RMAX) the adjacent cross-correlation function value in left and right and its corresponding coordinate point
(X is not denoted as it1,R1) and (X2,R2), this 3 points can fit the downward quadratic term curve of a recess, calculate the curve most
The corresponding coordinate of big value pointEnter step C2;
C2, core 0-7 are utilized respectively coordinate X' construction compensation terms at this time:And with as pre-echo stDot product is (right
Range cell is answered to be multiplied)With the current PRF echo s after being alignedt' and store, with this eight
Core respectively obtains the pulse echo after eight alignment, and respectively according to the current PRF echo s after alignmentt' calculate corresponding core
Partial reference signal:refI=st';So far the envelope alignment of current PRF echo is completed, and enters step D2;
D2, the partial reference signal update overall situation reference signal that core 0 is exported according to eight cores:
refg=γ8refg+γ7ref7+γ6ref6+γ5ref5+γ4ref4+γ3ref3+γ2ref2+γ1ref1+
ref0;0≤γ≤1;
Wherein, the partial reference signal that ref0 is calculated for core 0, the partial reference signal that ref1 is calculated for core 1, ref2
For the partial reference signal that core 2 calculates, ref3 is the partial reference signal that core 3 calculates, and ref4 is the part that core 4 calculates
Reference signal, ref5 are the partial reference signal that core 5 calculates, and ref6 is the partial reference signal that core 6 calculates, and ref7 is core
The 7 partial reference signals calculated;
Next pulse echo is taken out, according to new global reference signal, step A2 is repeated, until a frame echo data is complete
Complete alignment, the echo-signal after output alignment in portion.
And in the parallel processing of accumulation cross-correlation envelope alignment algorithm, it further includes and is changed using Fourier to global benchmark
Signal refgWith as pre-echo stCarry out interpolation processing:
It is assumed that the required precision of envelope alignment is 1/8 range cell, then need to global reference signal refgWith it is current
Echo st8 times of interpolation are carried out, the FFT of different points twice can be used in dsp to realize 8 times of interpolation, i.e.,:IFFT[FFT(st,
M), 8M], wherein, M is the range cell number contained by each pulse.First, it is single to reach 1/8 distance for envelope alignment required precision
Member, so the signal to be aligned to two is needed to carry out 8 times of interpolation, this is to believe the interpolation that length is N when being realized on DSP
Number N point FFT transform is to frequency domain, then mends 3.5N zero respectively at its both ends, and then 8N points IFFT becomes original time domain again.Secondly, profit
It is equivalent to the theory of frequency domain dot product with convolution, the time domain cross-correlation of pulse signal to be aligned and reference signal is by by two
Signal FFT transform is to frequency domain, then the two dot product result IFFT is switched back to time domain and is realized.
Meanwhile core 0-7 also improves cross-correlation function computational efficiency using Fast Fourier Transform (FFT) (FFT);
Calculate global reference signal refgWith as pre-echo stCross-correlation function R when, can be realized using 3 FFT, i.e.,:
R=IFFT [FFT (st).*FFT(refg)]。
In above-mentioned loop body, core 0 performs.If core 0 judges the mark place value of 8 core all as 2n+2, according to upper
State calculation formula update current base signal.Then the flag bit of all cores is increased ' 1 '.
Finally, it after core 0 detects that a frame data have all been aligned, exports to echo-signal thereafter, jumps out cycle
Body, core 0 perform:State flag bit is simultaneously reduced into initial value ' 0 ' in case the calculating of next frame data by 0 transmission data of core.
It is experimentally verified that, accumulation cross-correlation envelope alignment algorithm can not be realized at all on TS201, if in C6678 (dominant frequency
During for 1GHz) on sequentially calculate the accumulation cross-correlation envelope alignment of 256 pulse echos (each pulse has 1024 sampled points)
Algorithm needs 1268.1ms, and the method according to the invention then only needs 61.4ms, substantially increases computational efficiency.