SYSTEM AND METHOD FOR SENDING AND RECEIVING A DIGITAL SIGNAL OVER A RADIO PATHWAY FIELD OF THE INVENTION The invention relates to a method for transmitting and receiving a digital signal over a radio channel, said digital signal comprising information modulated with a constant or quasi-constant-amplitude modulation and the system enabling the implementation thereof. It is notably used in the field of digital telecommunications over radio channels, when the aim is to increase the efficiency of the transmitter by having it operate partially or totally at saturation. It applies notably for continuous-phase modulations, or CPM. It is useful in GSM mobile telephony systems that use a GMSK "Gaussian Minimum Shift Keying" modulation. More generally, it can be used for all the modulations that exhibit a constant or quasi-constant amplitude. Hereinafter in the description, the Applicant will use the following definitions: - the term "state" of a symbol is used to designate the representation in the complex plane of said symbol. - the term "constellation" of a complex sequence of symbols is used to designate the representation of said symbols in the complex plane. The term he(a(n), t) is the impulse response of a filter whose input is a(n) 8(t - n T) and is the frequency pulse associated with the symbol a(n). DISCUSSION ON PRIOR ART In some current telecommunications systems, the use of costant-amplitude modulations, for example phase or frequency modulations, is preferred, because the latter make it possible to maximize the range of said systems. In practice, the transmission power is constant and at maximum in this case. In the case of a continuous-phase modulation, there are two known advantages: • A reasonable spectral occupancy, which results in reduced interference between adjacent transmission channels, • A constant amplitude which makes it possible to use the output amplifiers of the transmitters at the maximum of their power, without having to worry too much about their amplitude linearity. This makes it possible to optimize the link budget, with a given average transmission power. Among these modulations, one of the most widely used, particularly in mobile radio telephony, is GMSK modulation. The latter has been adopted because of its frequency spectrum which exhibits a maximum decrease as a function of the deviation relative to the carrier frequency. It is a binary (and therefore two-state) modulation, and of differential type in that, when two successive bits to be transmitted are different (0/1 or 1/0), the carrier undergoes a total phase rotation of +TT/2 or of -TT/2 otherwise. The CPM modulations are in practice completely defined by a frequency pulse and by the modulation index h such that the average of the absolute value of the phase rotation is h n. Most of the time, and for the purposes of simplicity of implementation of the receiver, h is 1/2. However, the duration of the frequency pulse associated with a given bit is not limited to the duration of a bit. Thus, for the GMSK modulation mentioned previously, it ought to be infinite since, by definition, the Gaussian curve is of infinite length. In practice, a limit of finite duration (2 or 3 bits) is imposed, such that the performance degradation compared to the theoretical case is negligible. As long as the system is limited to the binary case, the receiver remains relatively simple. A theory expounded in the mid 1980s, explained in the publication "Exact and Approximate Construction of Digital Phase Modulations by Superposition of Amplitude Modulated Pulses (AMP)", Pierre. A. Laurent, IEEE Transactions on Communications, Vol. COM-34, No 2, February 1986, pp 150-160, showed that this type of modulation could be approximated by a conventional amplitude and phase modulation and therefore be demodulated by a receiver of low complexity. This is true only in the binary case (1 bit per symbol). At the current time, the needs in terms of useful bit rate have greatly increased, so much so that there is a desire to generalize the CPM modulations to more than two states: four states make it possible to convey not one bit per symbol but two, eight states 3 bits and 16 states 4 bits. Unfortunately, even in the case with 4 states, the receiver becomes much more complex than in the case with two states because the inter-symbol interference inherent in this kind of modulation considerably complicates the problem: the signal received for a given symbol depends on its state and on those of its neighbors and the number of configurations becomes so great that there is no simple way to decide on the value of said symbol. Moreover, the increase in bit rate also entails increasing the modulation speed with the attendant problem of the appearance of problems due to propagation: it may be that, at a given instant, the signal is received in direct sight of the transmitter, but with one or more delayed replicas (reflections on buildings, etc.) that have delays that are not inconsiderable in relation to the duration of a symbol, or even significantly greater. This further increases the complexity of the receiver. To the knowledge of the Applicant, in the case of a constant or quasi-costant-amplitude modulation, there is no transmission-reception system, of simple design, when the number of states envisaged is greater than 2. SUMMARY OF THE INVENTION The invention relates to a constant-amplitude and continuous-phase modulation method for modulating digital data and for demodulating said modulated signal, said data taking the form of symbols that can take a number M of states greater than 2, characterized in that it comprises at least the following steps: • recovering the symbols a(n) of digital data to be transmitted, being presented with a constant time interval T, • associating with each symbol a(n) a frequency pulse denoted he(a(π)), a frequency pulse of length L T, with L being an integer greater than or equal to 1, and whose form explicitly depends on the value of a(π), weighting the frequency pulse he(a(π), t) by the total phase variation associated with a(n), this total phase variation taking the form π(2a(n) -(M-1))/M with a(n) = 0, 1, ..., M-1, and for a symbol of value m, with 0<=m47dB M =4 states: SNRi > 33 dB M = 8 states: SNRi >28dB M = 16 states: SNRi > 27 dB Spectral characteristics The frequency spectra shown in the preceding section exhibit a strong and continuous decrease as the distance from the carrier increases, which is characteristic of a modulation based on Gaussian functions of which all the derivatives are continuous in any order. However, the fact that the frequency i pulses used exhibit time variations that are more rapid than a single Gaussian means that the frequency spectrum is wider. It is possible to limit the frequency spectrum to its central portion by filtering the signal just before the modulator. This is reflected in the fact that the amplitude of the signal transmitted is no longer strictly constant and this is reflected by a "peak factor" (ratio between the maximum power to be transmitted and the average power transmitted) changes between 1 and 1.5 dB if the filter is correctly chosen, which is entirely acceptable. The power amplifier has to be "slightly" linear and it is always possible to predictor the modulated signal to almost totally compensate its nonlinearity. By using the same filter in reception, the performance levels are scarcely degraded. Figure 8 represents the frequency spectrum after filtering (the peak factor is 1.3 dB) in the case with 8 states. It is possible to check that the pollution of the adjacent channels by a powerful interference will be much lower than previously. Figure 9 shows the constellation obtained, still with 8 states: the degradation compared to the case without filtering is insignificant, whereas the transmission and reception parameters are unchanged. If the filter used in transmission after the modulator and the filter used in reception before the matched filter are well characterized, the optimization of the different coefficients can be redone in order to take account of their presence. The method according to the invention notably offers the following advantages: On the transmission side Maximizing the efficiency of the transmission by virtue of a modulation with constant envelope. Maximizing the transmitted bit rate by virtue of a modulation with more than 2 states. Not requiring a transmitter with good linearity. On the reception side Having a receiver that is as simple as possible even for a modulation with more than two states (4, 8 or 16). Obtaining performance levels very close to theoretical performance levels. Being able to use the algorithms conventionally implemented for the so-called "linear" modulations, for example an equalizer of any type to fight against propagation issues. 1. A constant-amplitude and continuous-phase modulation method for modulating digital data and for demodulating said modulated signal, said data taking the form of symbols that can take a number M of states greater than 2, characterized in that at least the following steps: • recovering the symbols a(n) of digital data to be transmitted, being presented with a constant time interval T, • associating with each symbol a(n) a frequency pulse denoted he(a(n)), a frequency pulse of length L T, with L being an integer greater than or equal to 1, and whose form explicitly depends on the value of a(n), weighting the frequency pulse he(a(n), t) by the total phase variation associated with a(n), this total phase variation taking the form rc(2a(n) -(M-1))/M with a(n) = 0, 1, ..., M-1, and for a symbol of value m, with 0<=m