TW200423533A - Period and pulse width modulation - Google Patents

Abstract

A method of modulating analog or multi-valued digital signal, which generates three-valued pulse applied to switching power amplifier, controls the behavior of power transistor in the output stage. Said method wherein generating three-valued method while it concurrently modulate both period and pulse width, so called as period and pulse width modulation (Hereinafter referred to as the PPWM that stands for period and pulse width modulation). As generating three-valued pulse wave through digital system, it will arise quantization error due to the limit to base clock of digital system. Said method herein is a theory that mainly calculates analog or multi-valued digital signals, during a period of interval, with numeral integral, and that meanwhile seeks optimized period and pulse width, which will result in the minimum quantization error that remains appropriate switching frequency.

Description

200423533 V. Description of the invention (1) [Technical field to which the invention belongs] This invention proposes a method for modulating an analog signal or a multi-value digital signal to generate a binary pulse wave, which is mainly used in switching power amplifiers, especially in Sound Class D amplifier. [Prior technology] The so-called Pulse Width Modulation (hereinafter referred to as PWM) 'refers to a modulation analog signal or a multi-value digital signal to generate a two-vaiued pulse or a three-value valued) Pulse wave method. This analog signal or multi-value digital signal is hereinafter referred to as the input signal. The binary pulse wave means that the waveform size has only two constant values: the maximum value and the minimum value. If it is a unipolar input signal, the minimum value is 0. If it is a bipolar (bip〇1 ar) input signal, the minimum value is The value is negative. The three-valued pulse wave means that the size of the waveform is only the maximum value, the minimum value, and the three constant values, β β, which are small values, are negative values. The method of generating a binary pulse is usually applied to a bipolar input signal. The magnitudes of the maximum and minimum constant values are related to the actual system. The characteristics of the above two types of pulse waves are not directly related to the questions to be raised in this section ’, and both types of pulse waves will generate the topics to be mentioned in this section. The following content will mainly use binary pulse waves as an example. 、 When the input signal is positive, the size of the three-valued pulse wave generated by PWM is only the maximum value and 0 constant values. When the input signal is negative, the size of the three-valued pulse wave generated by PWM is only the minimum value. And two constant values. In both cases, the pulse wave changes the pulse width under a fixed pulse period.

Page 6 200423533 V. Description of the invention (2) ---, and its pulse width and pulse period ^ Different Duty Ratio can be referred to as the Duty Ratio. J input signal size. The main application areas of PWM are: B & / Tian JL · Device: An input signal is passed through a PWAjr to switch the power amplification wave to control the switching power amplification. Action to generate an output pulse, the output pulse, a switch of a power transistor, or a load with a low-pass filter function, and then a low-pass filter current, the output voltage or output current, ρ generates an output voltage or The size of the output signal is called the output signal, which is proportional to the input signal. The relationship between the two, and usually the pulse wave generated after the rate, the two pulses and power transistor amplification, the waveforms of the two are similar. The second aspect of the signal is a signal with a pulse size. §, both can be regarded as the same as PWM technology applied to switching power. In the past, the analog signal was used to compare the fixed period when the signal was already reported.-^ To use two comparators, the triangle wave and the input signal are like G ^ and the fixed period is violated. Signal & (S / N Rati〇), or various components such as Pete, will cause signal distortion, which will affect the accuracy of Duty Rati〇 and the output signal value. X further

Page 7 200423533

In the future, due to the development and maturity of digital technology, pure digital circuits can be used instead, and PWM can be implemented to reduce the above-mentioned signal-to-noise ratio and various signal distortion effects, and simplify the circuit. Although, in purely digital circuits, there is no need to generate ^ 3 fixed cycles, angle waves, and other analog components, such as comparators, etc., but the algorithm is basically the same as the principle of analog circuits. Produce 2 $! The value pulse should, in theory, be the same as an ideal analog circuit's binary pulse. However, pure digital circuits have a minimum clock, and the system is hereinafter referred to as the basic clock. Both the pulse period and pulse width must be integer multiples of the basic clock. Assuming that the pulse period is Tp㈣ and the period of the base pulse is Tbase, the resolution of the Duty Ratio is Tbase,

Tpwm, therefore, the three-valued pulse wave generated by a purely digital circuit cannot be exactly the same as the three-valued pulse wave generated by the analog circuit =, and the difference caused by the resolution is called the quantization error (Quantizati〇n Err〇) ^. Using purely digital circuits to implement PWM, the error 1 caused by the limitation of the basic clock is often not practical. For example, it is used in sound amplification: The pCMCD of general digital audio is "bits, It is used to represent the value of the bipolar sound signal in both positive and negative directions. The resolution is 1/6 5 5 3 6 and if the resolution in one direction is calculated, it is 1/3 2 768. Assume that there is a basic time for a pure digital PWM circuit. The pulse is 40MHz, and the frequency of the two-valued pulse wave is set to 40kHz, then each pulse wave period is 丨 00 basic clocks, so the resolution of its Duty Ratio is 1 / 1〇 ( ). Far lower than the resolution of pcM. Figure 1 A is an example of a sine wave input signal, with the & amplitude being the largest

Page 8 200423533 V. Description of the invention (4) 5% of the signal and the frequency is 3kHz. This sine wave is modulated by the pure digital pWM circuit to generate a three-valued pulse wave, and then passed through an appropriate low-pass filter. The output signal waveform appears as a stepped waveform as shown in Figure 1B. Figure 1c is the spectrum of the Fourier Transform of this three-valued pulse. In addition to a peak at 3kHz, this spectrum also has two smaller peaks at high frequencies. Noise caused by quantization errors. The smaller the input signal is, the greater the effect of noise on the output signal due to quantization errors on the output signal. Of course, you can increase the basic clock or increase the period of the three-valued pulse wave to improve the South resolution and reduce the quantization error. However, the more basic the clock is, the more advanced the process is used, and it involves the specifications of the entire system, which is a very complicated problem. The pulse period should not be increased too much, so as not to reduce the quality of the output number or affect the output. Signal bandwidth. Therefore, pure digital circuits are used to implement pwM technology, which is applied to switching power and amplified states, especially in sound amplifiers. How to improve the resolution to reduce the error of quantization is an important issue. There are many solutions to the resolution problem caused by the limitation of the basic clock of the digital system, as described below: The first method is to replace the pwM qPM and pwM qPM with pulse density modulation (hereinafter referred to as pDM). pWM is different

200423533 V. Description of the invention (5) Attach the pulse wave width at 疋 and change the pulse period to generate different Duty Rat 10 ′ _ to represent different input signal values. The input signal undergoes PDM modulation to produce a three-valued pulse. When the input signal is small, the resolution problem caused by the basic clock limitation can be avoided. However, the larger the input signal value is, the higher the value is. The shorter the wave period, that is, the three-valued pulse wave, the knife is switched to low 'and the switching frequency (sw 丨 7ch 丨 ng freqUfnCy) is switched from low to high. When the Duty Ratio, which represents the signal value, has the highest switching frequency around 0.5, the waveform of the three-valued pulse wave will be switched once for each basic clock cycle. When such a high switching frequency is used, a power amplifier of the output stage often results in a power loss of the output stage. Moreover, if the switching action of the power transistor Η f is at the basic clock cycle, the correct output pulse cannot be generated. It is necessary to put forward the message in advance: when implementing PDM, the signal is usually integrated, and then the value of the integrated operation is input to the quantizer to tune it into a three-value pulse. In the present invention, 3 The input signal of the internalizer is integrated for calculation, but the value of the integral calculation result / round method is different from the processing method of PDM, so it does not cause problems such as the switching frequency of the processing party. Method-The second method mentioned is to use closed-loop control technology to compensate for various errors including denier. However, to implement this technology, it is necessary to eliminate the wrong circuits or components, which will complicate the system. Pure number two: it's worth the analogy

Page 10 200423533 V. Description of the invention (6) Turn to note that one of the mainstream of switching power amplifiers using closed-loop control technology is called delta-sigma converter (ΔΣ converter). This type of replacement is mainly caused by PD M and closed-loop system are composed, so there is also the problem of switching frequency mentioned in Method 1. The third method is to change the output stage hardware circuit of the switching power amplifier. As shown in U.S. Patent Nos. 6, 4 9 2, 8 6 8 and 6, 5 3 5, 0 5 8, it is a "multi-reference voltage" method that provides two or more external power supplies to provide different resolutions. . The method of US Patent No. 5, 6 1 0, 5 5 3 is to use a boost (joost) or buck-boost (Buck-Bost) power supply circuit to increase its dynamic range to increase its resolution. The method of U.S. Patent No. 5, 4 2 2, 5 9 7 is to provide different power. Transistors' have different internal resistances and generate different output voltages to provide different resolutions.

All the techniques mentioned in the third method of Renmin ’s all require changing the hardware circuit of the output stage of the power amplifier, increasing the complexity of the system and increasing the cost. The present invention is directed to the above-mentioned problem, and a method of combining PDM and PWM is proposed, and a new method for generating pulse wave is proposed, which does not need to change the hardware circuit of the output stage of the power amplifier, but can effectively improve the resolution, reduce the quantization error, and make

Page 11 200423533 V. Description of the invention (7) The switching frequency of the output stage power transistor of the switching power amplifier is kept in a proper range, not too high or too low. [,] [Content] When implementing PWM with a pure digital system, as shown in the example in the previous section, due to the limitation of the basic clock, a quantization error due to the resolution will occur. In particular, when applied to a sound amplifier, this problem is more serious, because the minimum sound intensity change (jND, Just Noticeable Difference) that can be discerned by the human ear is about 0.8 dB (= 20 x 10 gl0 (sound intensity ratio)) 'That is, when the ratio of the two sound intensities is about 1.1, the human ear can tell. At the same ratio, the lower the sound intensity, the higher the resolution requirements. For example, in the pure digital PWM circuit in the example in the previous section, if the period of the ternary pulse wave is fixed at 100 basic clock cycles, the minimum change in the pulse width is 1 basic clock cycle. When the Duty Ratio is 20/100, the amount of change of 0.8dB roughly makes the Duty Ratio 22/100, which can still be represented by this digital system. When Du t y Ra t i 〇 is 2/1 0 0, the amount of change of 〇 8 d B roughly makes Duty Ratio 2.2 / 100, which cannot be expressed by this digital system. However, if the period of the ternary pulse is variable, it will be completely different.

200423533 Description of the invention Results. In the above example, if the pulse width is maintained at 2 basic clocks, but the pulse wave period is changed, Duty R ati 〇 may appear: 2/1 0 1 < 2/1 〇〇 < 2/99 ' In other words, the minimum allowable change is-with + 0 · 0 8 7dB, which is much smaller than JND's 0 · 8dB. This method of fixing the pulse width and changing the pulse period is called pulse wave density modulation (PDM, Pulse

Density Modulation), can improve the resolution of Duty Ratio, but 'this modulation method, the larger the input signal value, the more the period of the three-valued pulse wave & The frequency of switching to low and switching from low to high is increasing. When Duty represents the signal value

When Rft10 is around 0.5, it has the highest switching frequency, and the waveform of the three-valued pulse wave will be switched about every basic period. When such a high switching rate is applied to a switching power amplifier, it often causes output stage functions ^ ΐ, which results in too high switching loss. In addition, if the switching of the power transistor: the day-to-day interval is greater than the basic clock cycle, it will not be able to produce the correct output pulse and rotation signal. Too _ = purely digital circuit implements PWM. The three-valued pulses generated by it are fixed at the base period. However, due to the pulse-cycle PDM ', there is no problem that the switching frequency is too high. On the other hand, when generating a two-value pulse, the quantization error is smaller than that of PWM, but when the signal becomes larger, the switching rate increases with the increase in the Duty Ratio when it is approximately equal to 0.5. Causes too much switching loss, and even makes the switching power output pulse wave κ from the amplifier, and the power transistor can not operate normally.

Page 13 200423533

Considering the above-mentioned problems faced by PWM and PDM, the present invention proposes a method for generating a good pulse wave for the driver's car, which combines two modulation methods and simultaneously modifies the pulse wave period and pulse width to produce a high Resolution, and can maintain a three-value pulse with an appropriate switching frequency. I μ This method of modulating the input signal to generate a three-valued pulse wave is different from traditional PWM and PDM. It is hereinafter referred to as "pulse period and pulse width modulation and change" and is referred to as PPWM (Period and Pulse Width). Modulation). The principle of traditional P ^ VM 'Before generating a pulse wave, a fixed pulse wave period needs to be set first, and PPWM needs to adjust the pulse wave width and pulse wave period at the same time in order to minimize the error between Duty Rat i〇 and the input signal' It is difficult to determine the pulse period before determining the pulse width, so the traditional method of generating ρ ψ M is not suitable for PPWM. In addition, the principle of traditional PDM is to fix the pulse width 'and the pulse width of PPWM is not fixed, so the traditional method of generating pρM is not applicable to PPWM. A separate method of PPWM needs to be developed to tune the input signal into a three-value pulse. The following describes a method of PPWM, which tunes the input signal to a three-valued pulse wave, reduces the error between Du t y R a t i 〇 and the input signal value, and maintains an appropriate switching frequency. It should be noted that the following mainly discusses how to make the voltage value of the output signal proportional to the input signal value, and some applications mainly want to make the current value of the output signal proportional to the input signal value, but

200423533 V. Description of the invention (10) The following discussion is still applicable, as long as the discussion about voltage in the discussion is changed to current, and these changes are easily understood by experts in the technical field. The input signal of a switching power amplifier is usually an analog signal or a multi-valued digital signal, and the output voltage of an ordinary switching power amplifier output stage can only provide three types: positive supply voltage + V ss, negative supply voltage-V ss , With zero voltage. Therefore, at any instant, the voltage value of the output pulse of the amplifier is almost not equal to the voltage value required by the input signal.

However, the pulse width and pulse period of the output pulse can be adjusted so that within a small time interval, the integrated value obtained by integrating the voltage value of the output pulse of the switching power amplifier with time ^ and the time interval The integral value obtained by integrating the input signal and time is equal. The output pulse of the switching power amplifier is passed through a low-pass filter or directly connected to a load with low-pass filtering characteristics. The voltage value of the output signal is directly proportional to the input signal value. The following content describes how to generate the pulse period and pulse width of this output pulse:

Equation (1) shows that in the input signal, an arbitrary time point is taken as the time axis origin, and it is used as the starting point of a certain period of the output pulse wave that is desired to be generated. From this starting point, the integration of the input signal over time is performed between a pulse period T (at this time T has not yet been determined). At the same time as the desired output pulse

Page 15 200423533 V. Description of the invention (11) Integral. The input signal and the output pulse must have the same integral value ... · ⑴ JVin dt = JVppnm dt ............ ... τ τ ... .......... input signal

Vppwm ............ input tb pulse wave Τ ............ ............. A certain output pulse period Output pulses in this time interval have only two signal sizes. + Vss and 0, or -Vss and 〇, if the input signal size during this period is, then the output pulse is only + Vss and 〇, if it is input during this period. The value of fi is negative, so the output pulse wave has only two signal sizes of ~ V s s and 0, and Vss is usually the maximum voltage that the output power transistor of the switching power amplifier can raise. The following uses a positive round-in signal as an example to illustrate. At this time, the PPWM signal has only two signal values of + Vss and 0. If it is a negative input signal, the following derivation and description is still valid, as long as + Vss is changed, and some changes are easily understood by experts in the technical field. : Also, the wave width is DUty Rat i〇χ τ (At this time, the value of M ”Rati〇 is still determined). 'During this period, the output pulse value is + Vss, and the remaining time is 0, so' can be calculated. The integral value of the output pulse wave in this entire time interval is shown in formula (2):

Page 16 200423533 V. Description of the invention (12)

JVppwm T dt JOdt + JVss (1-DutyRatio) xT DutyRatio x

dt = DutyRatio TT x Vss .. (2) From the formulas (1) and (2), the relationship between the pulse width and the input signal can be calculated, and the number of pulse widths is defined as the ratio of the pulse width to the basic clock cycle. The expression is obtained as shown in (3):

_DutyRatiox Tbase

Definition: Pulse width number Vw Further, define a normalized input signal such as (4): Definition: Normalized input signal ··

Yin Vss (4)

Page 17 200423533 V. Description of the invention (13) Moreover, if the integral formula of the formula (3) is replaced by a numerical integral, the numerical integral of the normalized input signal is equal to the pulse width number, as shown in the formula (5).

T T

Tbase λ kxTbase Tbase

Nw = -CVnormal dt = \ [—- f ^ normal ^] ^ ^ Vnormal (k) ... (5)

Tbas4 tl TbaSe Tbase tl

Also: Numerical integration is performed once for each basic clock. K in (5) can only be an integer, and T needs to be an integer multiple of Tbase. So, define the number of pulse periods: T Tbase Number of pulse periods = impulse

Nw =

Vnormal (k)

In formula (6), every other basic clock, N p is incremented by 1 and a number of times of integration is performed. The value of the numerical integration result is N w and will change accordingly.

Page 18 200423533 V. Description of the invention (14) is not an integer. Therefore, according to formula (6), the ideal pulse width is usually not an integer multiple of the basic clock. However, when a digital system generates a pulse wave, the pulse width can only be an integer multiple of the basic clock. In order to minimize the error between the pulse width / basic clock period that can actually be generated and Nw, every time a numerical integration is performed, the value Nw of the integration result must be checked to see if it is closest to an integer. There are usually two adjacent integral values, an integer value and a value greater than this integer value. To determine the two integral values, which is the closest to the integer value, the algorithm will be too complicated, and regardless of the choice That factory, the error will not be too big. Therefore, the integral value obtained later in time can be fixed to facilitate the design rules as follows: When performing numerical integration, if a shape from decimal to integer appears, then Nw is the integral value closest to the integer. , This time ^, = is the "quasi-quantized time point" "and the integer part of Nw at this time point: for this: 'the pulse width number of the quasi-quantized time point", the Np at this time point is called this Number of pulse wave periods at quasi-normalized time points. " Because Vnormal is a normalized value, its absolute value is less than 1 after the numerical integration is performed by the parent. Therefore, when integrating at a value of 1 N w, the value of the smallest digit of the integer part from the decimal place to the integer ′ 仃 仃 will change if and only if it appears. So open every month ^ day inch ‘Nw’ Just monitor Nw

Page 19 200423533

Whether the quasi-quantized time point appears in the least significant bit of the integer part. If the value of 70 has changed, you can determine that the normalized input signal is different. Every time a quasi-quantized time point appears, such as the numerical integration that varies several times during the numerical integration process, as shown in the example in Figure 3. . Quantifiable time from several

The time point " is used to quantify the time point to choose one of them faithfully. And at this quantization time point, the conditions of the period of the pulse wave and the pulse width 'can not make the pulse wave period = the method, the nonlinear behavior during the actual hardware electrical operation should be considered to avoid the power transistor circuit Power-on crystals produce too large ::; wheel = signal quality has an effect 'and power grid, and the pulse wave period can not be too large. In addition-considering the above-mentioned factors of the output signal, you can use :: to avoid excessively large continuous waves. . Consider the above requirements. The following is a variety of different methods, which can be compared to the following simpler method: Two periods hereinafter hereinafter referred to as Ttypical, TtypiC £ this clock cycle Tbase < integer multiples, and Ttypical and Tbas

^ Ratio, set to Ntypical, to determine the pulse period: Be careful to wait for the typical on the time axis to look back for the last quasi-quantization time point, 砥 = this quasi-quantization time point is the quantization time point The number of pulses and the number of pulse periods are regarded as the number of pulse widths of a pulse period. As shown in Figure 3 ’When the numerical integration is performed to 30, the product?

Page 20 200423533 V. Description of the invention (16) The interval is Ttypi cal. In this interval, there are 3 quasi-quantized time points, as shown in 3 0 1, 3 0 2, 3 0 3, and 3 0 3 occurred last, so this point can be selected as the quantization time point to determine the number of pulse wave periods is 2 4 and the number of pulse widths is 3. According to the above, when Np is equal to Ntypical, look back to find the last occurrence, quantization time point, and select it as the quantization time point. The pulse number of this quantization time point is: the integer part of Nw when Np is equal to Ntypical. Serving. However, if the input signal is negative, N w will also be a negative value, so its integer part will also be negative, and the size of the output pulse can be determined by the positive and negative values of Nw. + Vss and 0 , Or -VSS and 0. Regarding the number of pulse wave periods at this quantization time point: a software variable or a hardware register can be set, which is hereinafter referred to as RegP. Regp is used to store the number of pulse wave periods. Whenever a quasi-quantization time point occurs, the Store Np at that time in RegP. When the next quasi-quantization time point occurs, store Np at that time = KRegP, and overwrite the previous value, so when Np is equal to Ttypical day guard, the value stored in Regp is the quantization time. The number of pulse wave periods at a point. After the pulse width and pulse period of a pulse wave are determined as above, the quantized time point of the I skin distance and the time point of NP = Ntypical. The time between the two times belongs to one of the next pulse wave period. Part 4 'The fractional part of donkey CNw' is the time of the selected quantized time point and Np = Ntypicai

Page 21 200423533 V. Description of the invention (I?) Point, between the two time points', normalizes the numerical integration value of the oversampling input signal. Therefore, subtract the integer part of Nw and subtract the number of cycles of this pulse from Np, that is, 'move the origin of the time axis to this quantization time point, and then repeat the above numerical integration to find the quasi-quantization time Point, select the quantization time point with Ntypi cal, determine the pulse wave period and other processes to generate the output pulse wave. In addition, there is an exception to be discussed: When Np is equal to Ntypical, if no quasi-quantization time occurs in this time interval Point, the above-mentioned method of selecting the quantization time point to determine the pulse period is not applicable. In this case, it means that the input signal in this time interval is very small, and its numerical integration result value is smaller than the minimum resolution that this pulse wave modulation method can provide. Therefore, N w can only be taken as an approximate value ′, and the integer closest to N w is the approximate value of N w. This approximate value is used as the pulse width number of this pulse wave, and N t y p i c a 1 is used as the pulse cycle number of this pulse wave. Then, reset N w and N p to zero, and start to produce the following _ According to the discussion of the above exception, the initial value of R eg P used to store the number of quasi-quantized pulse wave periods is set as ^ # value. The value stored in R eg RP can be used to determine what / any quasi-quantified time point between the temples on that day. It can also be set to Ntypical. * There is no quasi-quantization time point in this 2-valued integration interval, or the value stored in the hardware register is exactly ㈣. The method of ^ body variable judges whether the time interval is not accurate. Volume = square

Page 22 200423533 V. Description of the invention (18) The above method of generating pulse waves does not determine the position of the pulse in a pulse period. In fact, when the pulse frequency is much higher than the input signal bandwidth, the position of the pulse in a pulse period has little effect on the output signal. To simplify the algorithm, placing the pulse at the midpoint of the pulse period can produce good results. The above method generates the pulse width number and pulse wave period. If the pulse width number is negative, it means that the size of the pulse wave signal is -Vss and 0, and the true pulse width should take the absolute value of the pulse width number and the basic clock. Product of cycles. If the input signal is an early polarity signal, that is, the input signal has no difference between positive and negative, according to the above-mentioned method of generating pulse waves, the Nw obtained are all positive numbers, so the size of the output pulse wave generated is only + V ss and 0, The ternary pulse is degraded into a binary pulse.

Page 23 200423533

It is still a computer system with software and hardware. Designers can compose the system that implements this algorithm according to the marginal conditions and various kinds of data. In practical applications, non-technical factors, you should never rely on the hardware circuit of your device ^ ... a rod quasi-pulse oil T + · ”cattle and 糸, the first specification requirements, order £ {1 固 知 + pulse period D typical T + u η. 1 life TU sigh I this pulse period T base, and

The ratio of Ttypical to Tbase should be an integer

Ntypical: 勹 Positive number

Ntypical

Ttyp i ca 1 / Tbas e ............ (7) The purpose of the algorithm is to add ^ λ ^ waves. And this input signal == is converted into a binary value by the basic clock cycle via PPWM. According to the discussion in the previous section, it should be equal to = greater than the basic clock cycle. The input signal should be equal to r :: pa: lng), producing-over-taking radon, with a sampling period of 4 at the basic clock. See Figure 2. The experts of this oversampling mechanism are familiar with and easy to report a variety of implementation methods, which are implemented in this field, and are outside the scope of this invention.

In addition, a normalized operation converts a signal according to the previous section into a general signal for discussion. For this purpose, the normalized signal does not need to be numerically multiplied. For example, the points must be external circuits or 16 bits

Page 24 200423533 V. Description of the invention (20) 1 In terms of the numerical signal, if it is operated in a tired mouthpiece with more than 6 digits, it can be directly determined that the small 16 digits are all decimals, and when the value is Integral = When the result is rounded to the 17th digit, it is rounded from decimal to integer. , Σ The normalized oversampling signal passes through the circuit or microcomputer system according to the ppwM algorithm # to generate a data sequence, each data: there are two values, one is the number of pulse cycles, and the other is the pulse width This data sequence should pass a pulse wave generator to generate a true three-valued pulse wave. See the figure. There are also many implementation methods for this kind of pulse wave generator, which are well known and easy to implement by experts in this field, and are beyond the scope of this invention. The three-valued pulse wave generated by the pulse wave generator passes through a "power transistor control circuit" to control the switching action of the power transistor to generate an output pulse. After passing through a low-pass filter, an analog output signal is generated to drive The state of sound. Ling is shown in Figure 2. There are also a variety of implementation methods for this part of the circuit hardware, which are well known and easy to implement by experts in this field, and are beyond the scope of this invention. On the switching power amplifier, the relationship between the system and the PPWM generator is shown in Figure 2. The algorithm of the ρρ ″ generator will be described below. Raw

Page 25 V. Description of the invention (21) This algorithm includes the following pulse wave periodic variables or body registers, data:

^ v Another article or register ... RegP Knows the score variable or register: Nw Integral time variable or register: Np system, first basic & pulse count variable normalization oversampling / description / * n Like digital signal of sigma value · s (n) The above-mentioned normalized oversampling multi-valued a digital data.彳 σΚΚ, which is produced by the oversampling mechanism. The above-mentioned pulse direction variables or registers are described below. Please refer to Figure 4 for the algorithm flow. First, enter the algorithm flow, as shown in 401. 2. Reset Nw and Np to zero, as shown in 402. 3. Reset RegP to zero, as shown in 403. Fourth, Np is incremented by 1, n is incremented! , As shown in 40. 5. Perform numerical integration on S (η). ★ Miscellaneous ^ λΐ ^ 77 is also the action of accumulation, and the result is stored to Nw. As shown in 405. : 、、 When the accumulated result ’appears to be rounded to an integer, it is a π-quasi digitized time point ". If it is a quasi-quantization time point, proceed to step 7, otherwise skip to step 8. As shown in 4 0 6. 7. Store Np at this time to RegP, as shown in Figure 407. 8. Check whether the integration interval is equal to the standard period, that is, whether it is 200423533 5. Explanation of the invention (22)

Np = = Ntypical. If yes, go to step 9. If no, go back to step 4, as shown in 408. Nine, check whether there is no optimal quantization time in this integration interval. That is, whether RegP = = 0. If yes, go to step 10. If no, go to step 11, as shown in 409. 10. Generate pulse wave data. Because in the integration interval, no quasi-quantization time point 'indicates that the signal integration value of this interval is less than 1. set up·

Number of pulse wave periods = N t y p i c a 1 Number of pulse width = integer closest to Nw As shown in 4 1 0. Output the pulse wave data generated above to the pulse wave generator. 11. Prepare to generate the next pulse wave. Move the origin of the time axis to the end of the pulse wave this time and prepare to start the numerical integration of the next cycle. That is to reset the following two software changes or hardware registers:

Nw = 0;

Np = 0; skip to step 3. As shown in 4 1 1. 12. Generate pulse wave data. Select the last quasi-quantization time point as the quantization time point to determine the pulse period and pulse width, that is:

Number of pulse wave periods = R e g P; Number of pulse width = integer part of N w; as shown in 4 1 2. Output the pulse wave data generated above to the pulse wave generator 13. Prepare to generate the next pulse wave. Move the timeline origin to this time

Page 27 200423533 V. Description of the invention (23) The end point of the pulse wave is ready to start the numerical integration of the next cycle. That is, the values of the two registers will be adjusted as follows: N w = fractional part of N w;

Np = Np-RegP; skip back to step 3. As shown in 4 1 3.

As described above, it is a description of the entire algorithm. Figure 5A is the same input signal as Figure 1A. Fig. 5B is the waveform of Fig. 5A through PPWM to generate a three-valued pulse wave, and then an appropriate low-pass filter, the output waveform generated, compared with Fig. 1 B, it can be found that there is no step in Fig. 5 B Shaped waveform. Figure 5C is the frequency spectrum of the three-valued pulse wave after Fourier transform. Comparing it with Figure 1C, it can be seen that it has two sub-higher vertices at high frequencies than in Figure 1C.

Page 28 200423533 Simple illustration of the diagram [Simplified illustration of the diagram Figure 1 · Explained with an example, using a digital system to implement PWM, and

Figure 1 A Figure IB. Summerization error: An input signal waveform is a sine wave with an amplitude of five minutes and a period of 3 kHz. Hundreds of the Daxun number • The pure sine wave in Figure 1A is processed by a pure digital circuit, and the pulse wave ’is passed through an appropriate low-pass filter to generate a pWm shape. The digital system used here assumes that the wave

Figure 1C: Its pWM period is 400 kHz, so each two μ is a period of 100 system clocks, and its resolution is i / J 〇 〇. : The waveform of Fig. 1B is converted by Fourier transform. Fig. 2: Schematic diagram of PPWM applied to switching power amplifier system. Refers to the role of PPWM in the switching power amplifier system. Position 202 202 203 204 205 206 207 208 209 210 Input signal oversampling mechanism Normalized oversampling input signal PPWM generator contains pulse period and pulse Wave width data sequence pulse wave generator PPWM pulse wave power transistor control circuit bridge wheel output power transistor low-pass filter and load

200423533 Schematic illustration Figure 3: Method of generating PPWM. Take the example to illustrate the idea of PPWM: find the best quantization time point by numerical integration to reduce the error of output pulse

difference. 301: The best quantization time point is generated, the number of pulse rabbits =: 1, the number of pulse cycles 3 0 2: the number of periods = 8. Generate the best quantization time point, the number of pulses =: 2, the number of pulse cycles 3 0 3: the number of periods = 13. Generate the best quantization time point, the number of pulse rabbits =: 3, the number of pulse cycles 3 04: the number of periods = 24. Number of standard pulse wave periods = 25.0. Figure 3A: Representation-a normalized oversampling signal whose sampling period is equal to the basic clock of the digital system ▲ Figure 3B: The waveform obtained by performing numerical integration on the signal of Figure 3A Figure 3C: The pulse wave period and pulse width are determined by 5 in the waveform in Figure 3B to generate the pulse wave waveform. Figure 4: PPWM; Algorithm flow chart [, describing the implementation method of generating PPWM pulse data. 401 algorithm First step 402 Algorithm second step 403 Algorithm second step 404 Algorithm fourth step 405 Algorithm fifth step 406 Algorithm sixth step 407 Algorithm seventh step

Page 30 200423533 Schematic description 4 〇8: Algorithm eighth step 4 0 9: Algorithm ninth step 4 1 0: Algorithm tenth step 4 1 1: Algorithm eleventh step 4 1 2: Algorithm The twelfth step of the method 4 1 3: The thirteenth step of the algorithm Figure 5: Explained with an example, using a digital system to implement PPWM, the impact caused by the quantization error. Figure 5 A: —The test waveform is a sine wave with an amplitude of five percent of the maximum signal and a period of 3 kHz. Figure 5B: Waveform generated by processing the sine wave of Figure 5A through a digital system to generate a PPWM pulse, and then passing through a suitable low-pass filter. The digital system used here assumes that its basic clock is 40MHz and the standard pulse period of its PPWM is 40kHz. Fig. 5C: The spectrum obtained by transforming the waveform of Fig. 5B through Fourier transform.

Page 31

Claims (1)

200423533 VI. Scope of patent application 1. A first method for modulating a first input signal into a three-valued pulse wave. The first input signal is a normal analog signal or a multi-value digital signal. Multi-valued digital signal generated by η rma 1 ized and over-sampling. The three-valued pulse wave has only three constant values of 0, the maximum signal and the minimum signal. The first method is to perform numerical integration of a first input signal in a time interval, and use the value of the numerical integration to determine the pulse period and pulse width of the three-valued pulse in the time interval, and determine the next Time interval for numerical integration. 2. The first method described in item 1 of the scope of patent application includes at least the following three-step cycle: The first step: within a time interval, the value of the first input signal is executed as a unit integral. The second step: find a quantized time point in the process of performing numerical integration in the time interval, and set the time length from the start of the time interval to the quantized time point as the pulse of the three-valued pulse in the time interval The wave period is determined by the value of the numerical integration in the pulse period interval, which is determined by the pulse width of the three-valued pulse wave in the time interval. Step 3: Take the end of the pulse period as the starting point of the new time interval, subtract the integer part from the value of the numerical integration at the quantized time point, and set it as the initial value of the new numerical integration. In a step, the next numerical integration of the first digital signal is started in the next time interval to determine the next pulse width and pulse period. 200423533 VI. Scope of patent application. 3. According to the first method described in item 2 of the scope of patent application, the length of the time interval for performing the numerical integration is a constant value. 4. The quantization time point as described in item 2 of the scope of the patent application is a quasi-quantization time point in the process of numerical integration, whenever the value of the numerical integration changes from decimal to an integer. Select one of all quasi-quantization time points in the time interval to become a quantization time point. The time period from the start of the time interval to the quantized time point is set as the pulse period of the three-valued pulse in the time interval. Set the integer part of the value of the numerical integration in the pulse period interval as the pulse width number of the three-valued pulse wave in the time interval, and set the product of the absolute value of the pulse width number and the basic pulse wave period as its pulse wave. width. 5. Select one of the quantization time points from all quasi-quantization time points in the time interval as described in item 4 of the scope of the patent application. When there is no quasi-quantization time point in the time interval, the time interval is used. The end point is set to the quantized time point, the value of the numerical integration in the time period interval is used as the reference value, and the integer value closest to the reference value is reset to the value of the numerical integration. 6. In the case of rounding from decimal to integer as described in item 4 of the scope of patent application, determine whether the value of the numerical integral is rounded from decimal to integer. Page 33 200423533 6. The method of claiming patent scope is to monitor the minimum value of the integer part of the value of the numerical integration value. If the value of this bit is changed, that is, the value of the numerical integration value is rounded by decimals. In the case of an integer, the time point is set as a quasi-quantization time point. 7. As described in item 4 of the patent scope, the method of selecting one of all quasi-quantized day-to-day points in the time interval to become a quantized time point is a method that uses all quasi-quantized time points in the time interval. , Select the last quasi-quantization time point as the quantization time point. Set the pulse period from the time interval, ^ point to the quantized time point, and set the value of the numerical integration in the long pulse period interval as its pulse width number. The absolute value of ^ Ϊ ^ width number and the basic pulse The product of the wave period is set to its pulse width, and the pulse width described in item 7 of the patent range is used to set the pulse width unit. In the time interval, the basic clock period is used as the value to set the first digit. The signal performs numerical integration. The basic integral of the numerical integration is set to the pulse width number, and the product of the absolute value of the pulse width number and the T pulse period is set to the pulse width. 9, = Pulse period described in item 7 of the scope of the patent application, and the method of setting the pulse period = uses a software variable or a hardware register. The length of time between the start of the time interval and the time point is stored in the software variable or hardware Page 34 200423533 6. Register for patent application range. When the next quasi-quantization time point is found, the length of time from the beginning of the time interval to the next quasi-quantization time point is stored in the software variable or Hardware register and overwrite previous value. When the numerical integration of the time interval is completed, the software variable or the value stored in the hardware register is set to the number of pulse wave cycles. The product of the number of pulse wave periods and the basic clock period is set as the pulse wave period. 10. The software variable or hardware temporary storage as described in item 9 of the scope of patent application. 5 The initial value is set to the time length value of the time interval. 11 1. The software variable or hardware temporary register as described in item 9 of the scope of patent application, its initial value is set to 0 or any negative value. Page 35