Apparatus for converting, transmitting and reconverting a sampled signal
BACKGROUND OP THE INVENTION
This invention relates to an apparatus for converting, transmitting and reconverting a sampled signal and, more precisely, for converting a signal, preferably a speech signal, sampled in the time domain to a non-uniformly sampled signal in the frequency domain while excluding blocks of samples and, respectively, certain samples in blocks of samples in a transmitter, transmitting the signal thus non-uniformly sampled to a receiver, and reconverting the transmitted non-uniformly sampled signal to a uniformly sampled signal in the receiver.
PRIOR ART Apparatuses are known previoulsy, which comprise at the transmitter end a delay chain with a number of outputs, a time-frequency converter for converting short signal blocks of samples (for example eight samples) in the time domain to corresponding signal blocks of samples (eight samples) in the frequency domain, a logic unit for analys¬ ing the signals in the frequency domain and generating a control signal depending on the result of said analys¬ ing, and a change-over switch for excluding certain sampl¬ es (for example four) in blocks of samples in the frequen- cy domain.
The change-over switch is connected-in between the time- -frequency converter and the transmitting medium to the receiver end of the apparatus and is controlled by the control signal from a logic unit. The control signal in- eludes information of frequency ranges comprising samples with low signal amplitudes in relation to samples in rem¬ aining ranges. At a special embodiment, a second change- -over switch is located at the input at the transmitter end and excludes blocks of samples in the time domain prior to the conversion to the frequency domain in the time-frequency converter.
At the receiver end a frequency-time converter is locat¬ ed, which converts signal blocks of samples in the frequ¬ ency domain to corresponding signal blocks of samples in the time domain and also reconstructs the blocks of sampl- es previously excluded by said second change-over switch. Samples in these missing blocks of samples are reconstruct¬ ed in principle in the same way as samples in the other blocks by inverse conversion from frequency domain to time domain, but hereby they are formed, for example, as the mean value between corresponding samples in the adjacent two blocks or, in a more general and more applicable way, by weighting with suitable interpolation coefficients of corresponding samples in more than the very most iπ edi- ate blocks. The said inverse conversion of a block from frequency domain to time domain takes place by means of a delay chain with a number of outputs, a multiplication unit connected to the outputs of the delay chain and intended for weighted multiplication and subsequent addition of samples with coefficients stored in a memory, by controll¬ ing by means of control signals from the logic unit. T δ stored multiplication coefficients are so chosen that a desired reconversion to the time domain is obtained. Apparatuses of the aforesaid kind are known, for example through SE-PS 7905-^08-6.
Apparatuses of the kind described above per se have proved to operate well, and they utilize much of the redundancy comprised in speech signals, whereby good speech quality is obtained even when, for example, the data speed is reduced by the transmitting medium to 12-16 kbits/second.
It is possible by a further development of these apparat¬ uses to preserve to a still higher degree and in a simple way the redundancy comprised in speech signals, whereby a very good speech quality can be obtained at still lower data speeds.
SUMMARY OP THE INVENTION
The present invention has the object, thus, to modify the aforementioned previously known apparatuses so as in a simple and more efficient way, but to a still higher degree be capable of utilizing the redundancy comprised in speech signals. It was found by investigation that, when in- these apparatuses the transmitted frequency samples are; combined in pairs to an amplitude signal and a phase signal, these signals reflect the redundancy of the speech signals much better than frequency signals and, besides, are much more suitable for interpolation for obtaining previously excluded blocks of samples of the speech sign¬ als. This renders it possible to obtain a substantial data reduction at the transmission while maintaining a very good speech quality. For digitized signals the mathematic operations in available calculation units preferably can be carried out with so-called "look-up tables", because at short blocks of samples (for example eight samples) the number of quantization levels for transmitted frequency samples is low, for example eight levels at three-bits quantization.
The pattern for excluding certain blocks of samples can be varied, for example from every .second block to two subs≤- 3 quent blocks or even several blocks in one sequence corresponding to about the length of a pitch period of a speech signal, in order to obtain a greater data reduction.
The characterizing features of an apparatus designed according to the invention are apparent from the attached claims. BRIEP DESCRIPTION OF THE DRAWINGS
The invention is described in greater detail in the foll¬ owing, with reference to the accompanying drawings, in which Fig. 1 shows schematically an embodiment of an app¬ aratus according to the invention, and Fig. 2 shows sample configurations in eight di ferent points in the apparatus according to Fig. 1.
QMPI
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT The apparatus according to Fig. 1 comprises a transmitter unit and a receiver unit.
The transmitter unit comprises a delay chain Fa with a plurality of outputs, which is connected via a change- -over switch 02 (second change-over switch) to the input I of the apparatus, a time-frequency converter TFC conn¬ ected at the input end to said outputs, and a logic unit LO connected at its input end to the output of the time- -frequency converter TFC, and at its output end to the control . input of the change-over switch 02 and a control input of the change-over switch 01 and to a transmitting medium L. The change-over switch 01 (first change-over switch) is connected-in between the output of the frequen- cy converter TFC and the transmission πedium L.
The receiver end comprises in series from the transmitt¬ ing medium L a delay chain Fb, a calculating unit Bl, which converts frequency samples to samples of an amplit¬ ude signal and samples of a phase signal and has two out- puts d and, respectively, e, of which a first output d is connected to a delay chain Fl with subsequent multiplicat¬ ion unit Ml for weighted multiplication of samples with coefficients stored in a memory, and an addition unit SI, while a second output e is connected to a delay chain F2 with subsequent multiplication unit M2 of the same type as the unit Ml and an addition unit S2. After said units follow a second calculation unit B2 with two inputs, of which a first input f is connected to the output of the addition unit SI, and a second input g is connected to the output of the addition unit S2, a third delay chain F3, a third multiplication unit M3 for weighted multiplic¬ ation of samples, and an addition unit S3, the output of which is connected to the output U of the apparatus. The calculation unit B2 is capable to reconvert samples of the amplitude signal and phase signal to frequency
O PI
samples. The delay chains Fb,Pl,F2 and F3 comprise an equal number of delay elements.
Frequency samples are transmitted by a first line in the transmitting medium L via the first change-over switch 01 to the delay chain Fb, and by a second line control signals from the logic unit LO are transmitted to the multiplications units M1,M2, and from the unit M2 to the multiplication M3-
A uniformly sampled signal in the time domain ingoing to the input i is .' .converted one block after the other
(with for example eight samples per block) in the time- -frequency converter TFC to a signal in the frequency domain, and is analysed in the logic unit LO. The result is a control signal from the logic unit LO to the change-over switch 01, which excludes a certain pair of samples in the row of samples from the time-frequency converter TFC, viz. that pair of samples which relative to remaining samples in the block have low signal amplit¬ udes. The logic " unit LO also emits a control signal to the change-over switch 02, which excludes, for example, every second block or every second group of a number of blocks in series among those samples which lateron are reconstructed one block after the other in the tiπe- -frequency converter FTC. In Pig. 2, a, positions for a number of samples in an input signal, more precisely three blocks of samples designated I,II and III, with eight samples in each block, are shown. In Fig. 2, a', positions for samples in the blocks I and III and for excluded samples in block II are shown with regard to the point a'immediately after the change-over switch 02. In the time-frequency converter TFC blocks 'are con¬ verted one after the other to samples in the frequency domain according to prior art (see for example SE-PS 7905-408-6). Pig. 2, b, shows the position pattern for
samples in the point b at the output end of the converter TFC. In Fig. 2, c, positions for samples in the frequency domain are shown after certain samples have been excluded by the change-over switch 01 (also apparent from SE-PS 7905408-6), viz. samples Nos. 5 and 6 in the blocks I and III. The point C, thus, refers to the output end of the change-over switch 01.
In the calculation unit Bl frequency samples are convert¬ ed to amplitude samples A.;.»4-+τ and phase samples <-γ . . . by combination in pairs of adjacent frequency samples
U 1- and U1.+,_1, accordingw to the relations:
where i = 1,3,5 in the interval 1-p, where p is the number of samples (for example eight) in one block of samples.
As these sig&nal samp^les Aι.,ι.+,l and "01.,1.+,1. for speech signals have amplitudes, which vary with the time in a slow and regular way - slower and more regularly than corresponding frequency samples - they are highly adapt¬ ed for interpolation and, thereby, for the reconstruction of corresponding signal samples in excluded blocks. In Fig. 2, d and e, examples of the appearance of amplitudes of samples A 1.,1 -"*1 and -JJ l . jl . + in consecutive blocks in the point d and, respectively, e at the output end of the unit Bl are shown. The delay chains Fl and F2 together with the multiplications units Ml and M2 are units for interpolation for obtaining signal samples in excluded blocks in a manner known per se, i.e. by weighting with suitable multiplication coefficients of corresponding samples in adjacent blocks of samples. In Fig. 2, f and g, examples
•BURLA
OMPΓ
of the appearance of amplitudes of samples A. .+1 and
-P . . . are shown after such Interpolation in the point •J ι,ι+l t f and, respectively, g at- the output end of the addition units SI and S2.
In the calculation unit B2 a reconversion of amplitude samples and phase samples to frequency samples U^ and U. , is carried out according to the relations:
By means of the delay chain P3 and multiplication unit M3 frequency samples are reconverted to uniformly distr¬ ibuted samples in the time domain according to prior art, see SE-PS 7905408-6. As already mentioned, instead of every second block every second group of a number of blocks in series can be ex¬ cluded by the change-over switch 02. The number of blocks in such a group preferably is determined by possible periodicity of the speech signal. When the speech signal in the frequency domain has a periodicity with, for ex- ample, close to four blocks, preferably the size of such a group is chosen to be four blocks. At such a design, in the logic unit also a frequency analysis for investigating such periodicity is carried out, and the change-over switch 02 ;preferab1y ;i s pos i ti'onedi immedi ate-1y before or after the change-over switch 01, so that the signal to the logic unit is not affected by the function of the change-over switch 02. A method with a very good result
in respect of speech quality obtained in spite of the data reduction, in that half the signal information was excluded by the change-over switch 02, has proved to be to vary the size of excluded groups between one and at maximum seven blocks (of eight samples in the time dom¬ ain), depending on whether the periodicity shown is closest to 1,2,3,4,5,6 or 7 blocks.
-BURt T