Reduction of EMI through switching frequency dithering

[0002] This application claims the benefit of U.S. provisional application No. 60/154,690, filed Sep. 17, 1999 the disclosures of which are incorporated herein by reference in their entirety as if set forth at length.

[0003] All commercial electrical systems must meet specific Electro-Magnetic Interference (EMI) limits. EMI solutions are usually quite expensive. In some systems 30% of the whole cost is related to EMI/EMC (compatibility) for large filters, shielding, etc. In the Electric Power Steering (EPS) system the main contributor to RF noise is the Pulse Width Modulation (PWM) operation used for current control. PWM generates narrow-band noise in the form of a spike at every integer harmonic of the PWM switching frequency. The vehicle's radio is a serious victim to narrow-band noise, since the above harmonics can appear as distinct tones at the output of the radio. This invention converts the narrow-band noise into broadband noise and reduces the peak amplitude thereof by varying the switching frequency of the PWM within a small range.

[0004] The PWM harmonic spikes have an amplitude given by:

[0005] (At/T_s)·|(sin(πf_nt)/(πf_nt)|  (1)

[0006] where A is the amplitude of the pulses, t is the PWM “on” time, T_sis the PWM switching period and f_n, is an integer multiple of 1/T_s, the switching frequency, f_s. The spectrum is a series of spikes spaced apart by f_s, bounded by the modulus of a sine-shaped envelope as seen in FIG. 1.

[0007] A method of controlling the bandwidth of electromagnetic interference generated by a pulse modulation controlled automotive motor is disclosed. The method comprises providing a controller including at least one register thereof; the controller in signal communication with the motor for controlling the motor based upon operating parameters thereof; selecting a first clock frequency; selecting a first switching frequency; initializing the at least one register, establishing thereby a prescribed at least one number; selecting a first hopping period; based upon the prescribed at least one number of the at least one register, the first clock frequency and the first hopping frequency randomly modulating the switching frequency of the pulse modulation.

[0008] FIG. 1 is a representation of the spectrum of pulse width modulation showing spikes at harmonic frequencies and a modulus of a sine-shaped envelope.

[0009] FIG. 2 is a representation of the spectrum of pulse width modulation subject to frequency dithering.

[0010] FIG. 3 shows the resulting tone caused by fixed frequency pulse width modulation.

[0011] FIG. 4 shows the reduced volume “white noise” from 16 pulse width modulation frequencies selected pseudo-randomly every 250 μs.

[0012] FIG. 5 is a representation of the spectrum of a pure tone broadcast.

[0013] FIG. 6 is a schematic representation of an automotive motor in signal communication with a controller.

[0014] FIG. 7 is a logic diagram of a pseudo-random number generator.

[0015] FIG. 8 is a table of the output the pseudo-random number generator of FIG. 7 in one-to-one relation with an ordered list of pulse width modulation frequencies.

[0016] FIG. 9 is a first generalized flow diagram of the method of the present invention.

[0017] FIG. 10 is a second generalized flow diagram of the method of the present invention.

[0018] FIG. 11 is a third generalized flow diagram of the method of the present invention.

[0019] FIG. 12 is a generalized schematic representation of a system for effecting the pulse width modulation of a sinusoidally excited three phase load.

[0020] FIG. 13 is a representation of the narrow band noise of the motor current subject to fixed frequency pulse width modulation.

[0021] FIG. 14 is a representation of the broadband noise of the motor current subject to pulse width modulation frequency dithering.

[0022] The concept of PWM frequency “dithering” involves using a particular PWM switching frequency, f_s, only for a set number of PWM switching periods, T_s, before changing to another switching frequency, f_s, and making this “hop” periodically. Each switching frequency, f_s, used contributes unique harmonic content, but since it is not used constantly, the equivalent amplitude is much lower. The end result is a spectrum that is continuous and lower in peak amplitude.

[0023] The dithering process reduces the overall cost of the system. It does not require any additional components and is implemented through software. Allowing larger amplitude of RF noise while simultaneously decreasing the peak level of emissions reduces the cost needed for filtering and shielding.

[0024] There are numerous parameters involved in PWM frequency dithering, including switching frequency range, number of switching frequencies used, “hopping” frequency and the manner of frequency selection. The ideal case would encompass an infinite number of switching frequencies available (continuous distribution), changing thereamongst very quickly and selection therefrom at random. In this case the RF noise becomes pure broadband with amplitude lower than in the case of narrow band peaks.

[0025] Unfortunately, the ideal conditions are not realizable, so a compromise must be reached between system demands and broadband ideality. The EPS system is not capable of making truly random selections and it can only operate at a certain number of discrete switching frequencies. Changing the switching frequency very quickly can also cause throughput and stability problems. Generating a table of pseudo-random numbers within a range may approximate random selection of switching frequencies by ensuring that the sequence can not repeat fast enough to cause sub-modulation.

[0026] Simulation and theoretical calculations were used to predict results. However, the authoritative result is the audio output of the radio. Accurate models of coupling media and radio behavior can be derived, but the subjective nature of the human ear is not easily modeled. Therefore the real merit of any frequency dithering format is the final product, i.e., how it sounds. The following results were discovered:

[0027] 1) The hopping frequency, f_h, is the most important factor. The “sound” of the noise is highly dependent on the hopping frequencies. Fairly white sounding noise was achieved with a hopping period of less than 500 μs.

[0028] 2) Using as few as two switching frequencies can eliminate tones at the output of the radio (see #3 for the Δf between them).

[0029] 3) The range of frequencies from which to select should be determined by the IF bandwidth of the radio, the RF frequency range of the radio and the approximate PWM fundamental. The switching range should be big enough such that when multiplied by the harmonic number that corresponds to the radio's RF input, it exceeds the IF bandwidth. For example, for PWM near 20 kHz with an IF bandwidth of 9 kHz and a RF of 150 kHz (for European LW), the range of switching frequencies used should be greater than 9 kHz×20 kHz/150 kHz=1.2 kHz. Note that this is the minimum value that can be used and the maximum value is dependent on the stability of the system.

[0030] 4) Increasing the number of discrete switching frequencies used improves results, with diminishing returns for very large numbers.

[0031] FIG. 3 shows the resulting tone caused by fixed frequency PWM and FIG. 4 shows the reduced volume “white noise” from 16 switching frequencies, f_s, selected pseudo-randomly every 250 μs. The dithering process reduces the overall cost of the system. It does not require any additional components and is implemented through software. Allowing larger amplitude of RF noise while simultaneously decreasing the peak level of emissions reduces the cost needed for filtering and shielding.

[0032] Fixed frequency (standard) PWM generates harmonics at every integer multiple of the switching frequency as seen in Equation 1 and FIG. 1. Radio receivers and other electronic equipment are susceptible to this narrowband noise that is focused on specific frequencies. If an AM radio station broadcasts a pure tone, the resulting spectrum is shown in FIG. 5. Since the IF bandwidth of the radio is centered at the carrier frequency, f_c, and the smaller spikes (sidebands) both fit within the 9 kHz bandwidth, the radio decodes the 1 kHz tone. When a radiated harmonic from the pulse width modulation is close in frequency to a radio station, the radio demodulates it as an audible tone in exactly the same fashion.

[0033] Switching frequency dithering is an alternative method to Pulse Width Modulation (PWM). The goal is to spread the emitted RF energy over a larger frequency range (broader band) so that the resulting EMI is similar to white noise. Radio receivers are especially less susceptible to this type of EMI because the noise bandwidth is much larger than the IF bandwidth. For PWM, a switching period, T_s, contains “on” time and “off” time. The sum of these two times is the switching period, T_s. Current is controlled by varying the ratio of the on and off times, i.e., the duty ratio, D. Thus:

[0034] T_s=t_on+t_offand

[0035] D=t_on/T_s.  (3)

[0036] In standard PWM implementation, the switching period, T_s, is constant and only the duty ratio, D, varies. Switching frequency dithering is the process of changing T_s. This is done independent of the current control scheme. For example, constant current would mean varying T_swithout varying D; the on time and the off time would change together.

[0037] The ideal implementation of switching frequency dithering would involve switching frequencies, f_s, chosen at random from a large continuous distribution, hopping thereamongst as fast as possible. The resultant spectrum would become a pure broadband with the peak level lower than the narrowband peak level.

[0038] The hopping frequency, f_h, should be as high as possible. The only limiting factor is the switching frequency, f_s. The fastest possible implementation is to only use a switching frequency, f_s, for one switching period, T_s, before “hopping” a different one. For the hopping frequency, f_h, to be known exactly, the switching frequencies, f_s, must be chosen specifically as integer multiples of the hopping frequency f_h. EPS is constrained to 18 kHz<f_s<30 kHz by audio and thermal issues, so a high hopping frequency limits the number of switching frequencies that can be used. Using a flexible hopping frequency solves this problem. The hop to a new switching frequency, f_s, can only take place at the end of a switching period, T_s, so the new switching frequency, f_s, then begins. As an example, assume the hopping period (T_h=1/f_h) equals 500 μs, f_s1=19.5 kHz and f_s2=21 kHz. 19.5 kHz runs for 9 periods, totaling 462 μs. The 500 μs mark is reached during the tenth period, so that period finishes for a total of 513 μs. Pulse width modulation then happens at 21 kHz for 524 μs (11 periods). Note that the hopping frequency varies around 2 kHz. As an aside, this flexible hopping frequency also helps to eliminate modulating an audible tone in components of the system.

[0039] The ideal implementation is a continuous distribution of switching frequencies (N=∞). An electric power steering system (EPS) is capable of frequencies fitting Equation 2 where f_sis the switching frequency and n is an integer.

[0040] f_s=2.097 MHz/n84≦n≦116

[0041] 18.02 kHz≦f_s≦25.00 kHz  (2)

[0042] Below 18 kHz is the forbidden audio range and above 25 kHz is detrimental to thermal performance. This suggests that there are 32 potential switching frequencies, but limiting that to only about 10 through 20 improves system stability by limiting the total frequency range.

[0043] The lowest frequency band of interest for emissions determines the range of switching frequencies, f_s, that is required. The term broadband implies that the RF spectrum is continuous and without gaps. To accomplish this, the n^thharmonics of the PWM fundamental frequencies must overlap with the (n+1)^thharmonics in the frequency band of interest. The spectrum is fairly flat broadband above the frequency f_intin the above equation. For example: with a minimum switching frequency of 18 kHz and a maximum switching frequency of 21 kHz, the spectrum becomes flat (without gaps) above about 126 kHz. Thus,

[0044] f_int>f_max·f_min/(f_max−f_min)  (3)

[0045] The last parameter in the dithering process is the order in which the switching frequencies are used. Ideally this selection would be random, but the EPS controller cannot accomplish this. Fortunately, pseudo-random selection is adequate. Stepping through the switching frequencies in a given order causes sub-modulation in the audio range if the length of the series isn't sufficient. A pattern should take at least 0.5 ms before it repeats (f_h<2000 Hz). For this reason, a look-up table or pseudo-random generation is adequate.

[0046] In FIG. 6 the relevant parts of a system 100 for effecting the pulse width modulation frequency dithering are the EPS controller 102 and the EPS motor 104. Inside the controller 102 are power MOSFETs (shown at 500 in FIG. 12 by way of exemplification) used to switch the current to the motor 104. A Brushless Motor Drive (BMD) integrated circuit (IC) 106 controls these switches in either a voltage mode reset or a current mode reset manner. Inside the BMD IC 106 are two registers 106a, 106b. The first register 106ais for setting the Duty Ratio, D, and the second register 106bis for switching period, T_s. These registers 106a, 106bcontain numbers, N₁and N₂respectively which the IC must “count to” with a 2.097 MHz clock. At the beginning of a switching period, T_s, the proper MOSFETs 500 are activated based upon motor 104 position and desired direction of rotation. After the amount of time, Δt₁=N₁/f_clk, specified by the value in the Duty Ratio register 106bhas passed, these MOSFETs 500 are deactivated. The clock continues to run until the amount of time, Δt₂=N₂/f_clk, specified by the switching period register 106aexpires. At this point, the switching period, T_s, is reset and the MOSFETs 500 are reactivated. By changing these two registers 106a, 106bin proportion, the switching frequency, f_s, can be changed without changing the current level to the motor 104.

[0047] However, the Duty Ratio register 106bis not always the trigger that turns the MOSFETs 500 off. Sometimes the current level to the motor 104 triggers a reset. In this mode, a change of switching frequency, f_s, causes a small change in average (DC) current level. These changes induce audio modulation that can be heard from the motor 104. If the controller 102 changes between only a couple of switching frequencies, f_s, this would cause a tone of one-half the hopping frequency, f_h, to be heard in the motor 104. With several switching frequencies being used and a pseudo-random selection thereof, the noise from the motor 104 is somewhat white and is far less offensive.

[0048] The microprocessor of the controller 102 has serial communications with the BMD IC 106 periodically along a bus 102a. A new switching frequency command is sent to the BMD IC 106 at the end of the hopping period, T_h, for example, 500 μs; thus updating the switching frequency register 106a. When needed, the Duty Ratio register 106bis also updated. The next switching period, T_s, is at a new switching frequency, f_s.

[0049] Two frequency dithering methods are provided. Both use the same switching frequencies, f_s, and switch between them at the same hopping frequency, f_h. The difference between the methods is the demand that each places on the system memory and throughput. In the first dithering method a memory contains a list of switching frequencies that are loaded sequentially into the BMD IC 106. Each value in the list is a random choice from among several possibilities. After it is created, the list is stored permanently as part of the software. In the second dithering method, a shift register 200 is combined with a modulo-two summing function 202 to form a linear feedback register as seen in FIG. 7. Four of the register bits are XORed to form the input to an 8-bit register, creating a pseudo-random sequence that is 255 symbols long. Half of the contents of the shift register 200 form a 4-bit output 204 in one-to-one relation with 16 possible switching frequencies, f_s, as seen in the table in FIG. 8. The 16 switching frequency register commands are part of the software. In FIG. 8 the four leftmost columns represent the 4-bit output from the shift register 200. The column labeled “HEX” contains the hexadecimal values of switching frequencies supplied to the switching frequency register 106a. The column labeled “DEC” contains the decimal representation of the aforesaid hexadecimal values. The column labeled “Counter Cycles” contains the decimal representation of the last two digits of the hexadecimal values. The column labeled f_srepresents the switching frequency such that

[0050] 2.0970÷Counter Cycles=f_sMhz.  (4)

[0051] A generalized flow diagram 300 of the method of the present invention may be seen in FIG. 9. In box 302 the first register 106ais initialized to a first prescribed number, N₁, from a finite set of prescribed numbers and the second register 106bis initialized to a second prescribed number, N₂, from a finite set of prescribed numbers. In block 304 a first clock frequency, f_clk, is selected, a first hopping period, T_h, is selected and a first switching frequency, f_s1, is selected from a finite set of switching frequencies. In block 306, based upon the prescribed first and second numbers, N₁and N₂, the first clock frequency, f_clk, the first switching frequency, f_s1, and the first hopping period, T_h, the motor 104 is controlled by pulse width modulation at the first switching frequency, f_s1. In block 308, if the hopping period, T_h, has not elapsed, the motor 104 continues to be controlled by the pulse width modulation at the first switching frequency, f_{s, If Th}has elapsed, then at block 308a, if the switching period, T_s, has elapsed then in block 310, the first and second registers 106a, 106bare at least pseudo-randomly reinitialized to a third prescribed number, N₃, and a fourth prescribed number, N₄, in response to a command from the microprocessor 102 thereby at least pseudo-randomly changing the first switching frequency, f_s1, to a second switching frequency, f_s2, and thus controlling the motor 104 by pulse width modulation at the second switching frequency, f_s2. In block 308b, the switching period, T_s, is allowed to elapse before reinitializing the first and second registers 106a, 106b. The pulse width modulation control of the motor 104 at the new switching frequency, f_s2, continues until the hopping period, T_h, and switching period, T_s, have again elapsed and upon command from the microprocessor, the third prescribed number, N₃, and the fourth prescribed number, N₄, are thereby at least pseudo-randomly changed to N₅and N₆, thus, at least pseudo-randomly changing the second switching frequency, f_s2, to a third switching frequency, f_s3, thus controlling the motor 104 by pulse width modulation at the third switching frequency, f_s3. The aforedescribed pulse width modulation of the motor 104 at successive pseudo-random switching frequencies, f_si, continues in a cyclic fashion as the first and second registers 106a, 106bare reinitialized at the end of the hopping period, T_h, with the prescribed numbers, N_i.

[0052] A generalized flow diagram of box 306 of FIG. 9 is shown in FIG. 10. At box 314 the proper MOSFETs 500 are activated at the beginning of the switching period, T_s. At box 316 the amount of time, Δt₁=N₁/f_clk, specified by the value in the Duty Ratio register 106bis calculated. In box 318, at the end of Δt₁the aforesaid MOSFETs 500 are deactivated. At box 320, the amount of time, Δt₂=N₂/f_clk, specified by the switching period register 106ais calculated. At box 322, at the end of Δt₂, the switching period is reset and the MOSFETs 500 are reactivated.

[0053] The generalized flow diagram in FIG. 11 depicts a further embodiment of the method of the present invention. In box 402 the registers 106a, 106bare initialized to the first and second prescribed numbers, N₁and N₂. In box 404 the MOSFETs 500 are turned on. In box 406, if t_onof Equation 3 has elapsed then the MOSFETs 500 are turned off at box 408. In box 410 and box 412, if T_sand T_hhave elapsed, then the registers 106a, 106bare reinitialized at box 414.

[0054] The effectiveness of the present invention may be seen with reference to FIGS. 13 and 14. In FIG. 13 the traces designated by the reference numeral 600 and 600ashow relatively prominent spikes in signal amplitude at regular intervals of a fixed switching frequency, f_s, for two input frequency bandwidths. This is indicative of narrowband noise; whereas in FIG. 14 the traces designated by the reference numeral 700 and 700ashow a relatively smoother signal amplitude as a result of the dithering of the switching frequency, f_sat the same input frequencies. This is indicative of broadband noise.

[0055] While preferred embodiments have been shown and described, various modifications and substitutions may be made thereto without departing from the spirit and scope of the invention. Accordingly, it is to be understood that the present invention has been described by way of illustration only, and such illustrations and embodiments as have been disclosed herein are not to be construed as limiting the claims.