US 20030185405A1

(19) United States (12) Patent Application Publication (10) Pub. No.: US 2003/0185405 A1
Spencer et al.
(54) MODULATOR PROCESSING FOR A
PARAMETRIC SPEAKER SYSTEM
(52)

(43) Pub. Date:

Oct. 2, 2003

Us. 01. ............................................... .. 381/77; 381/79

(76) Inventors: Michael E. Spencer, Escondido, CA (US); James J. Croft III, PoWay, CA

(US); Joseph 0. Norris, Honolulu, HI

(57)

ABSTRACT

(Us)
Correspondence Address:
THORPE NORTH WESTERN 8180 SOUTH 700 EAST, SUITE 200 PO. BOX 1219

A parametric loudspeaker system using improved modula

tors to compensate for the non-linearity of the parametric process in air When driving the air at saturation levels and beloW saturation levels. The parametric loudspeaker uses a

SANDY, UT 84070 (US)

(21) Appl. No.: (22)

(63)

10/393,893
Mar. 21, 2003

pre-processed single sideband modulator that offers ideal linearity as characterized by square root pre-processed
double sideband modulators but With a loWer carrier fre

Filed:

quency and Without the Wide bandwidth requirements. By

eliminating some or all of the loWer sideband the carrier

Related US. Application Data

Continuation of application No. 09/384,084, ?led on Aug. 26, 1999, noW Pat. No. 6,584,205.
Publication Classi?cation
(51) Int. Cl.7 ............................. .. H04B 3/00; H04B 5/00

frequency can be reduced Without producing sideband fre

quencies in the audible range. LoWer operational frequen cies result in greater translation ef?ciency and greater output capability before reaching the saturation limit of air. A
pre-processor minimizes the effects of saturation limits for double sideband, truncated double sideband or single side

band processing to achieve superior output.

SS8 Channel Model

(without transducer)
ssa "L0
, -> Non -Linear Demod

Modulator

(Model)

120
_

Acoustic

I
l

Audio Output
tout(t) NLD cum)

553

mod(t)

Transducer

l y(t) Transducer

Modulator

>

Inverse W10)

I I

>

hm

> Non-Linear Demod > (Air Column)

Patent Application Publication

Oct. 2, 2003 Sheet 1 0f 12

US 2003/0185405 A1

Amplitude

l 1

11 16 21 26 31 36 4,1 46 51 56 61 65

Frequency Spectrum (kHz)

Fig. 1
(PRIOR Am)

Amplitude
f 4

6 11 16212631364145515661 56

III

Frequency Spectrum (kHz)

Fig. 2
(PRIOR Am)

Patent Application Publication

Oct. 2, 2003 Sheet 2 0f 12

US 2003/0185405 A1

E q

cl

11 1621263136414651 5661 66

Frequency Spectrum (kHz)

Fig. 3
(PRIOR ART)

%
5 _

R
(
E A

.....|||....
6 11 1621263136414651 5661 66

Frequency Spectrum (kHz)

Fig. 4
(PRIOR ARD

Patent Application Publication

Oct. 2, 2003 Sheet 3 0f 12

US 2003/0185405 A1

"O

G)

11 1621263136414651 5661 66

Frequency (kHz)

Fig. 5

5 11 16212631364146515661 66

Frequency (kHz)

Fig. 6

Patent Application Publication

15

Oct. 2, 2003 Sheet 4 0f 12

US 2003/0185405 A1

-10
w

i1 Ml!
I

I 1 I

A
6.05-04 8.0E-04

0.0E-04

ZOE-04

4.0E-04

time (s)

Fig. 7

Amplitude

L l

6 11 16212631 364146515661 66

Frequency (kHz)

Fig. 8

Patent Application Publication

Oct. 2, 2003 Sheet 5 0f 12

US 2003/0185405 A1

K 66
Ultrasonic

Non-Linear Demodulator (NLD)

Gain
AM
v De> square

Input to Air

Acoustic

Acoustic
Audio Output

I
->

Highpass
Filter >

modulator
30

f
32

F
34

()2

(iarmesDC)

\ Q40
38

36

Fig. 9A

DEemxopdulanti

2 .

1 _

100dB

120dB

140dB

160dB

180dB

Intensity

Fig. 98

Patent Application Publication

Oct. 2, 2003 Sheet 6 0f 12

US 2003/0185405 A1

/32
_ AM Demodulator

I Re
'"Pl"

Demodulated
M nitude

N Tranlslsrmer m 89/.l

H'Ib n

0MP"!

/ 42

l \
46

H \ 50 48

Fig. 1a

in
7 558 Channel Model

AUdIO

Acoustic

Input

55B

Transducer

NI-D

Audio Output

7_ Modulator

ha)
1

_ lvwzngzzzzyod
I

>
68

62

70

\ 64

\ 66

Fig. 1 1

/ 70
S58 Modulator
'

.I'mot
R 93'1
I

SSB

Moldulator "P"! /
72

Tranlslzrmer

HIb n

1R: Im
Fig. 12

Modulator Out ut
p
82

74

\80

Patent Application Publication

Oct. 2, 2003 Sheet 7 0f 12

US 2003/0185405 A1

ocmz q 02.3620
$30
8 3 \\

E8.203

8 k,AmE: bwoc?lE3m1Qz.03
mm m

Qkg m4ub2tnk ghaEo.sb-l7:10Q2S $3:35

ob2Li0-S.m>5cE3UQ
.mt 2
No
mmw Em
Ex En +

\ Qgmu4ab2cmwk

85:32 ntm ukaqmv

Patent Application Publication

Oct. 2, 2003 Sheet 8 0f 12

US 2003/0185405 A1

" 239:.0
_ _ _

_0Luwa|zouq EkmEAnuawQEkomS4c3bt2wA.030.2lI$9c~5bo,3E.m2Q

E M K u 2?: 3:530 i

E53va.-2H0P 3t0mu5a.nc:t

mF25w03:6

vosm2x 2o+3m82 .wA6?

m3m: .5 8:;

at I

_|I

-335. iu

Patent Application Publication

Oct. 2, 2003 Sheet 9 0f 12

US 2003/0185405 A1

0.3 92 03.86:30
0Eko4ua3b2mc0?
mm 5 o

BE S Li

E 3b2Eo:I17m0:Q2
I I I I I

an

Ex

F a It

I I
l I I
l

m350:m3.05
II_ l|Il_

N -

QLE\uOMUNWaHKCQEOU

B30E5.3 tmozuch

m352.m5 35 x35

touabmc

8m3:.593826?%

E3052i tuoabmch

E 2L2 A 03869,2

EkmvuacbomEx?

I I I l I I
I

I I I I I
I

I I I l l
J

Patent Application Publication

Oct. 2, 2003 Sheet 10 0f 12

US 2003/0185405 A1

E3 v

3Q B E SE23

E?
in 5 8 2 8.5036

.mt 2

SE83
E;
E0 B E 5.2MB
Ex

0.53 SQS

Patent Application Publication

Oct. 2, 2003 Sheet 11 0f 12

US 2003/0185405 A1

at F N

13tm0u5a.bcm

US 2003/0185405 A1

Oct. 2, 2003

MODULATOR PROCESSING FOR A PARAMETRIC SPEAKER SYSTEM

TECHNICAL FIELD

at the cost of in?nite system and transducer band Width. It is not practical to produce any device that has an in?nite

bandWidth capability. Further, the implementation of any

signi?cant bandWidth means that the inaudible ultrasonic

primary frequencies Will, on the loWer sideband, extend

[0001] This invention relates to parametric loudspeakers Which utilize the non-linearity of air When excited by high
frequency or ultrasonic Waves for reproducing frequencies in the audible range. In particular, this invention relates to

doWn into the audible range and cause neW distortion Which is at least as bad as the distortion eliminated by the in?nite

bandWidth square root pre-processing system.

signal processing and modulators for parametric loudspeak

ers.

[0006] In a typical application, the desired signal is ampli

tude modulated modulated on an ultrasonic carrier of

BACKGROUND ART

[0002] Aparametric array in air results from the introduc tion of sufficiently intense, audio modulated ultrasonic sig
nals into an air column. Self demodulation, or doWn-con

version, occurs along the air column resulting in an audible acoustic signal. This process occurs because of the knoWn physical principle that When tWo sound Waves With different frequencies are radiated simultaneously in the same
medium, a sound Wave having a Wave form including the

30 kHZ to 50 kHZ, then ampli?ed, and applied to an ultrasonic transducer. If the ultrasonic intensity is of suf? cient amplitude, the air column Will perform a demodulation or doWn-conversion over some length (the length depends, in part, on the carrier frequency and column shape). The prior art, such as US. Pat. No. 4,823,908 to Tanaka, et al., teaches that the modulation scheme to achieve parametric
audio output from an ultrasonic emission uses a double

sideband signal With a carrier frequency and sideband fre

quencies spaced on either side of it by the frequency difference corresponding to the audio frequencies of interest.

sum and difference of the tWo frequencies is produced by the non-linear interaction (parametric interaction) of the tWo
sound Waves. So, if the tWo original sound Waves are ultrasonic Waves and the difference betWeen them is selected to be an audio frequency, an audible sound is generated by

[0007] For example, When amplitude modulating a 6 kHZ

tone onto a 40 kHZ carrier, as shoWn in FIG. 1, sideband

frequencies are generated. FIG. 2 shoWs that the carrier

frequency (40 kHZ) is noW accompanied by a 34 kHZ

loWer-sideband and a 46 kHZ upper-sideband. Three com

the parametric interaction. HoWever, due to the non-lineari ties in the air column doWn-conversion process, distortion is introduced in the acoustic output. The distortion can be quite
severe and 30% or greater distortion may be present for a

ponents are noW present, 34 kHZ, 40 kHZ, and 46 kHZ Which gives a pure 6 kHZ envelope. As described previously, the 6 kHZ signal Would be square rooted before being used as the

moderate modulation level. LoWering the modulation level

loWers the distortion, but at the expense of both a loWer output volume and a loWer poWer ef?ciency.

modulation signal shoWn in FIG. 3. Using a spectrum produced by the square root function for the modulation signal of a 40 kHZ carrier generates the spectral components
shoWn in FIG. 4. Applying a square root function to the 6

[0003] In 1965, Berktay formulated that the secondary resultant output (audible sound) from a parametric loud
speaker is proportional to the second time derivative of the square of the modulation envelope. It Was shoWn by Berktay that the demodulated signal, p(t), in the far-?eld is propor
tional to the second time derivative of the modulation

kHZ signal produces in?nite harmonics, and the AM spec

trum has upper and loWer sideband frequencies that are also

in?nitely far from the carrier. It is infeasible to implement this type of system because of transducer bandWidth limi tations and similar problems.
[0008] In practice, the ?rst ?ve or six harmonics are

envelope squared.
32 2 (Equation 1)

enough to give a good approximation of the ideal square

rooted Wave. HoWever, even When the number of harmonics

PU) < WHWWW) l

is limited, the loW sideband frequencies still reach doWn into the audio range and create distortion. As in the foregoing example in FIGS. 1-4, the loWer-sideband frequencies that Would need to be emitted are 34, 28, 22, 16, 10 and 4 kHZ.

[0004] This is called Berktays far-?eld solution for a parametric acoustic array. Berktay looked at the far-?eld because the ultrasonic signals are no longer present there (by de?nition). The near-?eld demodulation produces the same audio signals, but there is also ultrasound present Which
must be included in a general solution. Since the near-?eld ultrasound isnt audible, it can be ignored and With this assumption, Berktays solution is valid in the near-?eld too.

This creates the problem that audible frequencies (16, 10,

and 4 kHZ) Will need to be emitted along With the ultrasonic
ones to make the desired modulation envelope.

[0005] The earliest use of this relationship for parametric loudspeakers in air Was a modulator design for parametric

[0009] Applying a square root function to the original signal reduces or eliminates the distortion in the demodu lated audio but it creates unWanted audible frequencies that are emitted. In the current state of the prior art, the only choice is betWeen high distortion (avoiding the square root function) or a Wide bandWidth requirement With less distor tion (using a square root function). Further, the square rooted

loudspeakers in 1985. This advancement included the appli

cation of a square root function to the modulation envelope.

signal for any given ultrasonic frequency is only valid for

loW level signals. As the ultrasonic poWer levels are

Using the square root function compensates for the natural squaring function Which distorts the envelope of the modu
lated sideband signal emitted to the air. Those skilled in the art have also shoWn that the square root double sideband signal can theoretically produce a loW distortion system but

increased to provide signi?cant audio output, the ideal envelope shifts from the square root of the signal to the audio signal itself (or 1 times the signal).

[0010] Another problem exhibited by parametric loud

speaker systems is that as the frequency and/or intensity of

US 2003/0185405 Al

Oct. 2, 2003

the ultrasonic sound Waves is increased to allow room for

lower sidebands and to achieve reasonable conversion levels in the audible range, the air can be driven into saturation. This means that the fundamental ultrasonic frequency is limited as energy is robbed from it to supply the harmonics. The level at Which the saturation problem appears is reduced

[0018] Yet another object of the present invention is to provide a parametric loudspeaker system to eliminate the
eXtended loWer sideband of a double sideband modulation

scheme used With parametric loudspeakers.

[0019] The presently preferred embodiment of the present

invention is a signal processor for a parametric loudspeaker system used in air. The signal processor has an audio signal input and a carrier frequency generator to produce a carrier frequency. The audio signal and the carrier frequency are miXed together by a modulator to produce a modulated signal With sideband frequencies Which are divergent from

6 dB for every octave the primary frequency is increased. In other Words, the poWer threshold at Which saturation appears, decreases as the frequency increases. Double side band signal systems used With parametric arrays must alWays be at least the bandWidth of the signal above any audible frequency (assuming a 20 kHZ bandWidth) and even
more if the distortion reducing square root function is used Which also demands an in?nite bandWidth.

the carrier frequency by the frequency value of the audio

signal. An error correction circuit is included to compensate

[0011] A further problem With prior art parametric loud

speakers is that they have a built in high pass ?lter charac

for the inherent squaring function distortion by modifying the modulated signal substantially Within said modulated signals bandWidth to approximate the ideal envelope signal.
The error correction circuit compares the modulated signal envelope to a calculated ideal square rooted audio signal and
generates an inverted error difference Which is then added

teristic such that the amplitude of the secondary signal (audio output) falls at 12 dB per octave for descending
frequencies. Because the loWer sideband of a double side

band system must be kept from producing output in the

audible range, the carrier frequency must be kept at least 20 kHZ above the audible upper limit for double sideband (DSB) and at the very least tWice that amount With a square

back into the modulated signal to correct for parametric

loudspeaker distortion. In one embodiment, an error correc

tion step adds neW errors but at a greatly reduced level. This

comparison and adding back of the error difference to the

rooted DSB. This range forces the carrier frequency up quite high. As a result, the saturation limit is easily reached and the overall ef?ciency of the system suffers.

original signal can be recursively implemented to decrease

the error to a desired level. Each level of recursive error correction tends to reduce the error by more than one half

[0012]

These excessive and undesirable types of distortion

preclude the practical or commercial use of the uncompen sated parametric arrays or even square-rooted compensation

and enough levels of recursive correction should be used to correct the distortion Without adding so many levels that more distortion is added. In alternative embodiments of the

schemes in high ?delity applications. Accordingly, it Would

be an improvement over the state of the art to provide a neW

present invention, the modulated signal can use forms Which

include but are not limited to a double sideband signal, a

method and system for pre-processing the audio signal

Which Would result in loWered distortion With a decreased

truncated double sideband signal or a single sideband signal.

[0020] These and other objects, features, advantages and

alternative aspects of the present invention Will become
apparent to those skilled in the art from a consideration of

bandWidth requirement for the ultrasonic parametric array

output. It Would also be desirable to use loWer primary frequencies Which are still above the audible range to

the folloWing detailed description taken in combination With

produce less saturation and attenuation.

OBJECTS AND SUMMARY OF THE INVENTION

the accompanying draWings.

BRIEF DESCRIPTION OF THE DRAWINGS

[0013] It is an object of the present invention to provide a method and apparatus to reduce the primary frequencies of

[0021]
[0022]

FIG. 1 shoWs a 6 kHZ tone;

FIG. 2 shoWs a 6 kHZ signal modulated onto a 40

a parametric loudspeaker system to thereby minimiZe air saturation and increase the conversion ef?ciency.
[0014] It is another object of the present invention to provide a parametric loudspeaker system Which corrects distortion Without increasing the required bandWidth to
reduce the distortion.

kHZ carrier signal;

[0023] FIG. 3 shoWs the frequency spectrum of a 6 kHZ signal after the application of the square root function;

[0024]

FIG. 4 shoWs a 6 kHZ signal after application of

the square root function and modulation onto a 40 kHZ

carrier signal;
[0025] FIG. 5 shoWs the modulation of a 6 kHZ single sideband signal modulated onto a 40 kHZ carrier;

[0015] It is another object of the present invention to provide a method and system for pre-processing an audio signal that Will result in loWer distortion and better repro
duction of an acoustic audio signal for a parametric array

[0026]

FIG. 6 is a 5 kHZ and 6 kHZ single sideband signal

output.
[0016] Another object of the present invention is to pro
vide a parametric loudspeaker system that uses a double sideband modulated signal Which has a truncated loWer sideband.

modulated onto a 40 kHZ carrier;

[0027] FIG. 7 is the ideal envelope shape With the square root function applied Which Would result from the single sideband spectrum;
[0028]
[0029]

[0017]

It is another object of the present invention to

FIG. 8 shoWs the insertion of arti?cial sideband

FIG. 9A is a non-linear demodulator model for a

provide a parametric loudspeaker system using pre-pro

cessed single sideband modulation With reduced bandWidth

frequencies to model the ideal envelope shape of FIG. 7;

requirements.

parametric array in air;

US 2003/0185405 A1

Oct. 2, 2003

[0030] FIG. 9B shows a graph of the damping function used for the demodulation exponent;
[0031] FIG. 10 is an AM demodulator based on a Hilbert

audio signal With the square root applied. The ideal signal is the unmodulated audio signal after it has been offset by a

positive DC (direct current) voltage equal in magnitude but

opposite to its maximum negative peak value and then square rooted. As mentioned, this is because the audio signal that demodulates in a parametric speaker is proportional to
the square of the modulation envelope. Therefore, an enve

transformer;
[0032] FIG. 11 is a single sideband channel model;
[0033] FIG. 12 is a more detailed vieW of the single sideband modulator in FIG. 11;

lope that is proportional to the square root of the incoming audio Will be converted back to the original audio signal
upon demodulation in the medium.

[0034]
sator;
[0035]

FIG. 13 is a modulation side distortion compen

[0043]
FIG. 14 is a ?rst order baseband distortion com
FIG. 15 is a Nth order audio distortion compen

The frequency response of the ultrasonic trans

pensator;

ducer to be used is also taken into account in the comparison. In other Words, a correction is also added Which accounts for

[0036]
sator;

the distortion created by the transducer (i.e. speaker) When

it emits the ultrasonic signals. Before the envelopes are compared, the modulated signals bandWidth or spectrum is multiplied by the actual frequency response curve of the transducer/ampli?er combination. This ensures that the comparison betWeen the ideal envelope and the modulated signal envelope is valid because the modulated signal enve

[0037] FIG. 16 shoWs a Nth order audio distortion com pensator as a cascade of distortion models;

[0038]

FIG. 17 is a SSB channel model implemented as

the magnitude squared of the Hilbert transformed input;

[0039] FIG. 18 is an AM channel model using an AM modulator.
DISCLOSURE OF THE INVENTION

lope Will be altered by the transducer/ampli?er When it is

emitted. An embodiment using truncated double side band

(TDSB) may be partially truncated by the transducers

high-pass frequency response, or the modulation scheme
itself may also truncate the TDSB before it reaches the transducer. This makes it possible to use a simple DSB multiplier unit to generate a conventional DSB signal and a ?lter and the transducer to convert the DSB signal into a

[0040] Reference Will noW be made to the draWings in Which the various elements of the present invention Will be

given numerical designations and in Which the invention

Will be discussed so as to enable one skilled in the art to

TDSB signal.

make and use the invention. It is to be understood that the

[0044] The modulated signal envelope is then compared or

subtracted from the ideal square rooted signal. This gives a neW signal that represents the error. This neW signal is then

folloWing description is only exemplary of certain embodi

ments of the present invention, and should not be vieWed as

narroWing the claims Which folloW.

inverted (in phase or in sign) and summed With the original incoming audio signal just ahead of the modulation step.
This serves to alter the resulting envelope so that it is a closer

[0041] This invention is a signal processing apparatus and method, implemented either digitally or in analog, Which signi?cantly reduces the audible distortion of a parametric array in air. Within this invention, multiple signal processing steps are performed. The input side of the processor(s)
accepts a line-level signal from an audio source such as a CD

match to the ideal envelope. A signi?cant feature of the

present invention is the error terms that are calculated and

player. In the digital implementation, an analog audio signal

Will ?rst be digitiZed or a direct digital input may be received. The ?rst step in the invention multiplies the

then added back into the audio signal are alWays Within the audio bandWidth of the original audio signal and no extra bandWidth is required. In another embodiment of the inven tion, the primary distortion correction occurs Within the audio signal but some of the distortion correction terms may be outside of the audio signal if the added terms do not

incoming audio signal by a higher ultrasonic carrier fre

quency to create a modulated signal. In other Words, the

produce signi?cant distortion.

[0045] Adding the calculated error correction does not correct the envelope in one step, because the envelopes

carrier frequency is modulated by the incoming audio signal

to generate a conventional single sideband (SSB) or double

sideband (DSB) signal. The carrier signal is generated by a

local oscillator set at the desired frequency. Note that in a

frequency spectrum is not proportional to the incoming audio frequencies only. The envelope is proportional to the
square root of the sum of the squares of the modulation

multi-channel system (stereo, for example) only one oscil

lator is preferably used so that all channels have exactly the same carrier frequency. This modulation may produce either

spectrum and the modulation spectrum shifted by 90 degrees. In other Words, each introduced correction fre
quency produces other smaller error frequencies that must also be corrected. Accordingly, the error correction is pref erably done recursively a number of times until the SSB,
DSB or TDSB envelope error versus the ideal signal is Within a desired small amount. The number of recursive

a single-sideband (upper sidebands only) (SSB) multiplied

With a carrier signal, or a double sideband (DSB) multiplied With a carrier signal. A truncated double sideband (TDSB) signal may also be produced in the invention, Where the loWer sidebands of a double sideband (DSB) signal are sharply truncated by a ?lter so nearly all of the frequencies passed are above the carrier.

[0042] Next, the calculated envelope of the modulated signal is compared to the calculated ideal audio signal
With the square root applied. This comparison uses the modulated carrier envelope to compare against the ideal

steps Will depend on the desired amount of distortion reduction and on the practical limits of the processor. The modulated signal is then output to an ampli?er and ulti mately to the ultrasonic transducer Where it is emitted into
the air or some other medium. The ultrasonic Waves then

demodulate into the original audio signal according to Berktays solution.

US 2003/0185405 A1

Oct. 2, 2003

[0046] Each recursive step reduces the total harmonic distortion (THD) error percentage by at least one-half, With the actual amount depending on the incoming spectrum and
the modulation method chosen. The number of recursive

level (SPL), so a higher SPL can be used. The higher the SPL

used, the greater the conversion ef?ciency (betWeen ultra sonic and audio). In fact, the amplitude of the audio signal
generated is proportional to the square of the ultrasonic SPL. In other Words, the gain of the system increases With increasing drive levels, until the saturation limit is reached. The saturation limit is increased by loWering the carrier frequency. Third, a loWer carrier frequency increases the volume velocity available to the system and therefore increases the available output in the audible range.

steps is dependent upon the processing poWer available and

the desired level of correction. Generally, a half-doZen iterations or less of the recursion process produces the

desired distortion correction. The processing poWer required

for this level of correction in real-time is loW and could be implemented on an inexpensive DSP chip, or equivalent hardWare. As previously described, a carrier modulated by a square rooted audio signal has in?nite bandWidth and cannot

[0050] For example, the single sideband (SSB) method is

used to speci?cally decrease the carrier frequency as far as

be emitted accurately by any knoWn means. Using this method makes it possible to approximate the ideal envelope Without requiring the substantially increased bandWidth that is otherWise required. It should be recogniZed that error
correction could be performed With only one level of error

possible Which maximiZes the ef?ciency of the ultrasonic

to-audio conversion. With a loWer frequency saturation carrier, higher saturation levels can be achieved because the

correction if desired. Analog circuitry could also be used

instead of a digital or softWare implementation of the invention.

acoustic saturation limit is higher With longer acoustic Wavelengths. The ideal envelope can be created using only
the upper sidebands of a carrier modulated by an audio

signal.
[0051] There are several additional advantages to using

[0047] In a digital embodiment of the invention, the modulated signal Which is an ultrasonic frequency Would usually be converted back into analog form before ampli? cation. A high sampling rate is needed for a faithful digital to analog conversion in the output stage. For example, if the SSB carrier frequency Was 35 kHZ, and the input audio bandWidth Was 20 kHZ (the nominal value), the output signal
Would have a spectrum from 35 kHZ to 55 kHZ. A sampling rate of 96 kHZ or higher Would be a good choice. The standard 44.1 kHZ tends to be insuf?cient for Wideband audio. In contrast, certain applications for speech could use

single sideband (SSB) amplitude modulation. These bene?ts

include: eliminating the need to apply the square root function to the audio, reducing the transducer bandWidth

requirements, and greater ultrasonic conversion ef?ciency

because loWer carrier frequencies are used. In order to make the ideal envelope to create a single audio tone, SSB Without a square root applied gives exactly the same envelope as

offsetting, applying the square root, re-offsetting, and using

double sideband (DSB) AM. To create a 6 kHZ tone When

loWer sampling rates. Further, the output signal for the digital implementation is at line level. This signal Would be
input to an ultrasonic ampli?er Which Would in turn drive the

using SSB the folloWing spectra are needed as shoWn in FIG. 5. This is much simpler than the double sideband (DSB) of FIG. 4 or FIG. 2. The envelope and the demodu lated audio Which results from the spectra in FIG. 5 is

transducer. Again, the demodulated signal is proportional to the square of the modulation envelope. At higher ultrasonic
amplitudes Where saturation comes into play, the demodu lated audio begins to be proportional to the envelope itself,
not its square. This can be taken into account in the error

exactly What is generated by the in?nite spectra in FIG. 4,

if it Were possible to implement the hardWare required to generate FIG. 4. Thus, applying a square root and the
associated offsets can be eliminated With the SSB method.

correction compensator if the ?nal drive level is knoWn. For example, if the ampli?er and the signal processor Were
integrated, the error correction scheme could vary With the

This is a great advantage because the distortion and the logic required are reduced.
[0052] Of course as the complexity of the audio signal increases, the SSB method becomes less of a perfect sub stitute for the full square root method. HoWever, by arti?

poWer output in relation to the the ampli?er settings. Varying

the error correction With the poWer output is described in more detail later. For simpler systems, the square of the envelope can be used as the demodulation model With good results.

cially adding extra upper sideband components Within the

signal bandWidth, SSB can be made to match the ideal

envelope very closely. FIG. 6 shoWs the reproduction of

simultaneous 5 kHZ and 6 kHZ tones. This SSB spectra Would normally look like What is shoWn in FIG. 6. The ideal

[0048]

By using a SSB or a TDSB system, the carrier

frequency and modulated signal frequencies can be loWered Without Worrying about the loWer sidebands Which Would otherWise be emitted in the audible range (i.e. audible

envelope shape With the square root applied is shoWn in

FIG. 7 Which is the Waveform that Would result from the SSB spectrum in FIG. 6. It is important to note that the

distortion). The carrier frequency and modulated signal

frequencies can be loWered so they are close to the upper

amplitude of the SSB signal does not alWays match the desired envelope shape. HoWever, if another upper sideband
component is arti?cially inserted, a much better ?t can be
achieved. FIG. 8 shoWs Where a neW component is inserted

limit of the audible range. In this invention, close is de?ned,

as close to the upper limit of the audible range as possible

Without producing signi?cant distortion and Where the car rier signal and sidebands are inaudible.
[0049] A loWer carrier frequency alloWs for better con version efficiency in three Ways. First, the attenuation rate of
the ultrasound is loWer so the effective ultrasonic beam

for this example so that the SSB signal more closely

represents the ideal Wave form of FIG. 7. The neW fre

quency component in this case is 41 kHZ. Adding in addi tional frequencies is a very simpli?ed version of the error
correction that Was described above. In each case Where

additional frequencies are added, the neW sideband fre

length is longer, and the available energy isnt absorbed by the medium quite so quickly. Second, the shock formation (saturation) length is increased for a given sound pressure

quency is equal to the carrier plus the difference betWeen the tWo upper sidebands. In this example, the carrier is 40 kHZ and the dominant sideband frequencies are 5 kHZ and 6 kHZ

US 2003/0185405 A1

Oct. 2, 2003

so the arti?cial sideband is 41 kHZ, and no extra bandwidth

It should be realiZed based on this disclosure that applying

is required When inserting this neW component. Essentially, the tWo frequencies With dominant magnitudes can alWays
be used to determine the location of the neW sideband.

a damping function is similar to pre-processing the signal by applying the square root at loWer signal poWer and then
increasing the square root function to 1 as the poWer of the

[0053] Using a SSB or TDSB scheme is advantageous because it more ideally matches the amplitude output of a typical ultrasonic transducer above and beloW its resonant frequency. For example, the carrier in an SSB or TDSB arrangement Would be placed at the fundamental resonant

signal and saturation increase. This function Which interpo

lates the square root up to one can be modeled as either a

linear function, quadratic (n2) function or a cubic (n3)

function.

[0057]

FIG. 10 expands the AM demodulator block of

frequency of the transducer for maximum speaker output

levels, and the upper sideband frequencies Would fall on the upper side of the resonant peak Where the transducer oper ates efficiently. Many transducers Work Well above the

FIG. 9A With the ideal instantaneous AM demodulator based on the Hilbert transformer. An ultrasonic signal is received at the input 42 and passed to the Hilbert transformer 46. The Hilbert transformer 46 is a linear ?lter that simply

resonance frequency, and poorly beloW this peak frequency.

[0054] NoW a more detailed embodiment of the invention
Which uses a recursive error correction scheme Will be

shifts the phase of any input tone by 90 degrees Without affecting its amplitude. For example, an input of b cos(u)t) is
transformed to an output of b sin (out). The magnitude block
48 computes the square root of the sum of the squares of the

discussed and block diagrams of the invention Will be described. Although the preferred TDSB method is dis
cussed, SSB or DSB are also thoroughly described. In the

real and imaginary inputs, thus extracting the signals

instantaneous amplitude Which provides a demodulated out

invention, a distortion compensator is positioned after the modulator to cancel ?rst-order distortion products. A ?rst
order base-band compensator is used Which can also be recursively extended to an Nth order distortion compensator.

put 50.
[0058] An SSB channel model 60 Will noW be described

Which models an uncompensated parametric array system

that uses a SS B modulator 70. Referring noW to FIG. 11,

The base-band compensators pre-distort the audio signal

prior to modulation. When the ?rst order distortion correc tion is applied it creates smaller distortion terms Which are then corrected in the next level of recursion. Signi?cant

a single sideband (SSB) channel model 60 is constructed by

adding a SSB modulator 70 and the ultrasonic transducer response 64 in front of the non-linear air column demodu lator (NLD) 66. An audio input 62 enters the SSB channel model and an acoustic audio output 69 model is produced.

distortion improvements have been shoWn using the Nth

order compensator With various modulation schemes.

[0055]

The ?rst component of the invention models the

The ultrasonic transducer 64 (i.e. speaker) is modeled by the linear ?lter, h(t) and is typically bandpass in nature. The
NLD details are given in the description of FIG. 9A.

non-linear demodulation Which occurs in the air column of

a parametric speaker. This relationship must be modeled to provide a proper approximation of the distortion Which is
needed to produce the correct acoustic sound Wave. The

[0059] The SSB modulator 70 is expanded in FIG. 12 and speci?cally performs upper sideband modulation With car
rier feed-through. It is assumed that there is no DC term

second derivative function in Berktays solution (Equation 1) presents a linear distortion that may be compensated for by passing the audio signal through a double integrator prior to subsequent processing and modulation. Since the focus
here is to control the non-linear distortion component, the

present in the modulator input 72. The modulator input 72 is

received and the Hilbert transformer 74 is used to derive the

complex analytic signal having real RE and imaginary parts

IM prior to the summing node 76. Unlike a real signal, With

its negative frequency components equal to the conjugate of

its positive frequencies, it can be shoWn that the analytic signal has no negative frequency components. The modu

derivative Which can be handled by simple equaliZation techniques Will be dropped from this discussion. FIG. 9A
shoWs a block diagram representation of a non-linear demodulator Which does not model the second derivative.
Ultrasonic acoustic Waves 30 are emitted into the air Which

lator 78 modulates the analytic signal With ejt, and right

shifts its spectrum by (no. The constant, 1 is added to the signal path in the summing node 76 to cause some carrier signal to pass through. Taking the real part 80 restores the

performs a demodulation function modeled by the AM

demodulator 32. Since an audio signal cant contain a DC term, a high-pass ?lter 30 has been added to the model to remove the DC component from the output of the squarer block 32. A gain constant, a is included at 38 for scaling purposes and an acoustic audio output is then generated 40. The air column demodulator in the ?gure is referred to as the
non-linear demodulator or NLD.

negative frequency components of the signal. In effect, the single sideband modulator shifts the audio spectrum right by
(no and adds a carrier tone at (no.

[0060] To summariZe the SSB method, the distortion of a SSB modulator With discrete tone input signals can be

[0056]

In an alternative embodiment of the invention, the

squaring function in the non-linear demodulator uses an

exponent Which decreases as the intensity of the ultrasonic signal increases. The demodulation exponent of this inven
tion can increase from 1/2 to 1 in a smooth curved fashion or

reduced by this invention. The distortion products have frequencies that are equal to the differences of the primary input signals. Additionally, the distortion tones have a loWer amplitude than the primary input tones if the modulation index is less than one (amplitude of the carrier signal is

greater than the peak modulated signal amplitude). So, if

additional input tones are injected at the distortion frequen cies it completely cancels these ?rst-order distortion prod
ucts. The result is that second-order distortion products are introduced at the additional tone difference frequencies.

it can be linearly interpolated from 1/2 to 1. Increasing the exponent, models the air saturation that takes place as the poWer of the ultrasonic signal increases. FIG. 9B shoWs the damping function of the demodulation exponent With respect to the intensity in decibels of the ultrasonic signals.

HoWever, the amplitude of the secondary distortion products is signi?cantly less than the original distortion amplitude,

US 2003/0185405 A1

Oct. 2, 2003

resulting in an overall improvement of distortion ?gures.

Application of additional canceling tones in a recursive manner further improves output distortion.

[0061] Injecting Weak tones at the distortion frequencies improves the overall distortion. Distortion-tone injection Works by observing the amplitude of the distortion and injecting a tone With the same amplitude and opposite phase.
This Works because the SSB channel model passes input

114, and then the original audio input 112 is subtracted from the estimated distorted signal 114 leaving the distortion dist(t), 118. This distortion is scaled by the parameter c, (0<c 1), 120 and subtracted 122 from the original audio

input 112, resulting in the ?rst-order pre-distorted audio

signal, x1(t) at 124. The cancellation parameter, c controls the percentage of the ?rst-order distortion that is canceled.

tones Without signi?cant amplitude or phase modi?cation, and superposition (summation) applies at the acoustic output
facilitating the cancellation. This assumes a unity gain transducer model.

[0067] Since the SSB channel model produces distortion products With frequencies equal to differences of the inputs,
no frequency expansion occurs at any node in the system.

Thus, if the input bandWidth is limited to 20 kHZ, then the

bandWidth of the distortion, dist(t), and pre-distorted signal,

x1(t) are also limited to 20 kHZ. The single sideband

[0062] In the preferred embodiment of this invention compensating for the distortion of broad-band signals, not
just tones, is desired and the distortion components of a

modulator simply right shifts (translates) the spectrum of

x1(t) and adds a carrier. Therefore, the bandWidth of mod(t) is also limited to 20 kHZ (although the center frequency is high). The main implication of this is that the actual trans ducer bandWidth need only be 20 kHZ Wide and the inverse

general, Wide-band input signal must be estimated. Estimat ing the distortion in the Wide-band modulated signal Will
noW be described.

?lter, h_1(t) need only perform inversion over the 20 kHZ

band of interest. One of the bene?ts of this system is that dif?cult transducer responses may be dealt With easier.

[0063]

This invention uses a modulation-side distortion

compensator, shoWn in FIG. 13, that predicts, then cancels

the ?rst-order distortion components after the SSB modu

lator. By analyZing the SSB channel model in real time, the

distortion component can be estimated as shoWn in FIG. 13.

Assume initially that h(t) is unity or 1. The audio input 92

is SSB modulated 70 and then demodulated With the NLD 66 and transducer model 64, to derive an estimate of the

[0068] The ?rst-order compensator of FIG. 14 is easily extendable to higher order compensators by the recursive application of additional stages. The Nth order distortion compensator is shoWn in FIG. 15. Here, the pre-distorted signal, x1(t) is used as the input to another distortion
compensator, and so on, until the desired order is reached.

output of the uncompensated parametric array 96, or

outd(t)=x(t)+d(t), Where x(t) is the desired input signal and d(t) is the distortion. By subtracting the input signal from
outd(t) in the summation node 99, We are left With the

FIG. 15 shoWs that the audio distortion is recursively estimated using SSB Channel Models. A portion of the

estimated distortion signal is subtracted from the pre-dis torted input by each level of recursion, thus reducing dis
tortion in the acoustic output. There is a point of diminishing
returns Where no further improvement is attained as the

distortion products d(t), 100. Next, We frequency shift the distortion products up With the SSB (suppressed carrier)
modulator 90 to get the modulation error signal e(t), 102.
The error signal has no carrier signal present because it Was

compensator recursion levels are increased, especially for a

high modulation index.

[0069] The Nth order distortion compensator may be also
vieWed as the cascade of distortion models subtracted from the audio input as shoWn in FIG. 16. It can be shoWn that the alternate con?guration of the Nth order distortion com

removed in the SSB suppressed carrier modulator 90. This error signal 102 is subtracted from the main modulator output 106 in the adder 104 to mitigate the ?rst order distortion products in the ?nal acoustic output.

[0064] This compensator also Works for the case the h(t) is approximately unity. The system may be modi?ed to handle an arbitrary transducer response by including a
transducer inverse model. This is not detailed here because the base-band distortion compensator discussed beloW is the

pensator of FIG. 16 simpli?es the block diagram of FIG. 15 and gives additional insight into the operation of compen
sator. From the block diagram in FIG. 15, We see that the

pre-distorted input signals are given by

xhl(t)=x;(t)c;(M(t))xO(t)) i=0, 1, 2, . . . , N-1 (Equation 2)

most preferred embodiment.

[0065] NoW, base-band distortion compensators Will be

addressed. Another method of distortion abatement is to

[0070] Where M(') is the channel model and xO(t) is de?ned as the input; x0(t)=X(t). Next, de?ne the distortion
generator system, D(') as the difference betWeen the channel

subtract the distortion products from the main modulator input as detailed in FIG. 14. This is knoWn in the invention
as a ?rst-order distortion compensator. Here, the transducer

model output and its input,

(Equation 3) D(Xi(l))=M(Xi(l))-Xi(l) [0071] Let the cancellation parameters be unity, ci=1 for
all i. Note that D(xi(t)) is the distortion or error signal

response, h(t) is ignored in the SSB channel model 110 because its inverse is applied prior to the actual transducer.

The cascade of h_1(t) and h(t) is approximately unity (at

least over the frequency range of interest) so tout(t)Emod(t). The audio distortion is estimated using the SSB Channel Model. A portion of the estimated distortion signal is sub tracted from the audio signal, thus reducing distortion in the acoustic output.

generated by the non-linear plant. It is Zero only When the

plant is distortion free. Combining equations (2) and (3), We get an alternative expression for the pre-distorted signals,
x;+1(t)=x0(t)D(x;(t)) i=0, 1, 2, . . . , N-1 (Equation 4)

[0072] Equation 4 is depicted in FIG. 16 and shoWs that

the Nth order distortion compensator may vieWed as the

[0066]

In this embodiment of the system, the SSB channel

cascade of distortion models subtracted from the original

model 110 is used to derive an estimate of the ?rst order

audio input.
[0073] The SSB channel model may simpli?ed Which
creates a more efficient implementation for the distortion

distortion products dist(t). The distortion is estimated by

using the SSB Channel model 110 to estimate the distortion

US 2003/0185405 A1

Oct. 2, 2003

compensators. FIG. 17 shows that the Hilbert transformer based AM demodulator Works for any carrier frequency,

[0080]

For example, if a transducer With a 40 dB roll-off

including uuO=0. Making this substitution alloWs the SSB

Channel Model to be realiZed as the magnitude squared of

the Hilbert transformed input.

[0074] Since the SSB channel model is used as part of the distortion controller, an ef?cient implementation is desir able. The SSB channel model (excluding the transducer

from 40 kHZ to 50 kHZ is used, and tWo equal amplitude tones, 1 kHZ and 9 kHZ, are input to an uncompensated system, resulting in a 35 dB amplitude mismatch. A 6th order compensator Will reduce the amplitude mismatch to

only 3 dB. Using both phase and amplitude compensation

gives better overall results With only a second order com

pensator.
[0081] Considerable simpli?cation of the AM channel model may be performed if the transducer response is unity
over the complete AM modulation spectrum, or a unity

response) is expanded in the top 150 of FIG. 17. One of the properties of the AM demodulator using the Hilbert trans form is that it Works independent of the carrier frequency of the modulator. This includes uuO=0. Making this substitution eliminates the need to do the ?rst Hilbert transform 160, saving a signi?cant amount of circuitry or DSP (digital signal processor) resources, depending on the hardWare

response over both upper and loWer sideband frequencies, (a 40 kHZ bandWidth). A unity response is generally not the
case because Wide-band transducers are dif?cult to build.

implementation 170.
[0075] The basic principle of the Nth order recursive
distortion compensator also Works With an amplitude modu lator. The channel model must be rede?ned to include the AM modulator as shoWn in FIG. 18. Substituting the AM channel model into the base-band compensator results in an effective distortion control system that avoids the complexi ties of the single sideband modulator. Unlike the SSB case,
bandWidth expansion is an issue in the AM case because an

[0082] Another useful simpli?cation is to loWer the carrier frequency of the AM modulator in the AM channel model and shift doWn the frequency response of the ?lter g(t), so
that it is in the correct position relative to the carrier. The ?nal modulator remains at the desired carrier frequency. Only the carrier frequencies of modulators in the AM channel models are reduced. These changes preserve the

input/output relationship of the AM channel model, but loWer the maximum signal frequency to tWice the system
bandWidth (e.g. maximum frequency of 40 kHZ for a 20 kHZ

AM modulator has the property of doubling the signals

bandWidth. The Nth order distortion compensator of FIG. 15 is modi?ed for the AM case by substituting in the AM channel model from FIG. 18 and the AM modulator in place of the SSB modulator.
[0076] The ultrasonic transducer Will typically cut off or attenuate a portion of the loWer sideband of the AM fre quency spectrum. For this reason, the ?lter g(t), is required in the AM channel model to simulate this attenuation. Minimum requirements for this ?lter is that it be linear phase ?lter and have a bandpass characteristic similar to the actual transducer used in the system. The ?lter should be modeled
as the cascade of a compensation ?lter and the transducer

system). This simpli?es a DSP based implementation by reducing the sampling rate.
[0083] It is to be understood that the above-described arrangements are only illustrative of the application of the

principles of the present invention. Numerous modi?cations and alternative arrangements may be devised by those skilled in the art Without departing from the spirit and scope
of the present invention. The appended claims are intended
to cover such modi?cations and arrangements.

What is claimed is:

1. A signal processor for a parametric loudspeaker system,

comprising:
at least one carrier frequency generator to produce a

?lter, that is (Equation 5)

carrier frequency;
a modulator Which receives at least one audio signal and modulates the at least one audio signal onto the carrier

[0077] Where is the convolution operator, hcomp(t) is the compensation ?lter, and h(t) is the transducer response.
[0078] There are tWo alternative approaches to designing

frequency to produce a modulated signal, Wherein the

at least one audio signal is converted to sideband

the compensation ?lter. The ?rst option is to choose hcomp(t)

as the approximate inverse of the transducer response h(r). This choice Will ?atten out the amplitude response of the cascade g(t), and lineariZe the phase. In this case, g(t) is a model of the cascade of the transducer inverse and the transducer ?lters as in the bottom portion of FIG. 15. This

frequencies Which are divergent from the carrier fre quency by the frequency value of the at least one audio

signal;
an error correction compensator coupled to the modulator

to compensate for inherent to parametric demodulation

is the preferred option because very loW order (?rst-order)

distortion controllers are effective.

distortion by modifying, substantially Within the modu lated signals bandWidth, the modulated signal to approximate the ideal audio signal Which should be output by the system.
2. The signal processor as in claim 1 Wherein the error

[0079] The second option is to compensate only for the phase of the transducer model With hcomp(t). Gain variations With frequency Will be present in the cascade g(t). In this case, for example, a pair of equal amplitude tones may emerge at the output With different amplitudes. This ampli
tude error Will be treated as distortion. The effect of the Nth

correction compensator adjusts for the inherent parametric demodulation distortion by comparing the modulated signal
With a reference signal Which models parametric demodu lation distortion, and thereby generates an inverted error difference to add back into the modulated signal substan tially Within the modulated signals bandWidth to correct for distortion.
3. The signal processor as in claim 2 Wherein the error

order compensator Will equaliZe the amplitude difference

betWeen the tWo tones and improve the distortion. HoWever,

performance suffers When compared to using phase and

amplitude compensation.

correction compensator further comprises: