Beruflich Dokumente
Kultur Dokumente
RIGHT
POSITIVE
+
RIGHT
N EGATIVE
L OW-PASS
FILTER
L OW-PASS
FILTER
LEFT
R IGHT
PWM D/A CONVERTER
PERIPHERAL AUDIO CIRCUIT
Figure 2. Blo ck diag ra m of a PWM / MASH digit al-to-analog c onv erter.
A digi ta l fi n i t e i m p u lse r esp on se (F IR) fil t e r p rod uces 18-bi t d ata from a 16-bi t
in p ut s am ple aft e r 4-tim es ove r s am pli ng. T h e noise s ha pe r s th e n conve rt th is 18-
bi t d ata int o an 11-s t ep q uant ize d format for th e PWM aft e r 8-tim es ove r s am pli ng.
T h e PWM sys t e m is ope rat ed at 768 tim es th e origi na l s am pli ng fr eq u e ncy (33.868
M Hz). If it we r e to actua lly do a 1-bi t conve r sion of 16-bi t sig na ls, 65,536 p ulses
wou ld be n eeded to r ep r ese nt e ach am pli tu de. Howeve r, th is wou ld r eq u i r e th e
con ve rt e r to ope rat e at speeds i n excess of 2.98 G H zfa s t e r than th e curr e nt ly
a v a il a ble bip ol ar tran sis tor t echn ology. T h is r es tra i nt i m poses th e r eq u i r e m e nt
that th e 18-bi t d ata be r ed uced to th e 11-s t ep out p ut . I n p ra ct ice th e MAS H
conve rt er can only be con side red a 3.5-bi t conve rt er.
T h e secon d low-bi t con ve r sion t echn iq u e by Ph ili ps is kn ow n a s p u lse-de n si t y
mod ulat ion (P DM) or Bi t s tr e am conve r sion. I n th is t echn iq u e, th e de n sity rat io of
th e sig n of th e p u lses is r el at ed to th e or igi na l 16-bi t d ata wor d. T h e P DM
con ve rt e r is a tru e 1-bi t t echn ology. T h is sig na l r ep r ese ntat ion ma y not see m
i mm edi at ely obviou s. A si m ple mode l h elps ill u s trat e w hat is ha ppe n i n g
4
. If a
13
4X OVERSAM PLIN G
(FIR FILTER)
1-BIT D/A
(SW IT CHED CAPA CIT OR
N ET WOR K)
SEC OND -ORDE R
N OISE SHAPER
lig ht is on, th e n th e room is b rig ht ly li t; if th e lig ht switch is off, th e room is d ark .
B ut if th e sw i tch is cycle d ra pidly on an d off, an i nt e rm edi at e i nt e n si t y can be
cr e at ed . T h e s am ple d ata from th e decode r chip is fir s t p a ssed to a low-p a ss non-
r ecur sive 4-tim es ove r s am pli ng F IR int e r polat ion fil t e r. T his type of fil t e r yiel ds
hig h e r q ua li ty becau se it is p ha se-li n e ar . Fir s t-or de r noise s ha ping is pe rform ed
by th e accumu l at or of th e mu l t iplie r i n th e fil t e r . T h e secon d fil t e r i n g s ta ge
co n s i s t s of a 32 - t i m es ove r s am p li n g li n e ar i nt e r pol at o r an d a 2- t i m es
ove r s am pli n g s am ple an d hold ci rcu i t . At th is s ta ge, a 352 kH z digi ta l di th e r
sig na l at -20 dB is a dded to th e s am ple sig na l. T h is r ed uces non li n e ar i t ie s
in d uced by q uant iz at ion noise. At th is point , th e tota l ove r s am pli ng is 256 -tim es
an d th e d ata wor d ha s incr e a sed to 17-bi t s. T h e d ata is th e n fed at a fr eq u e ncy of
11.2896 M Hz int o th e secon d or de r noise s ha pe r. T h e noise s ha pe r r ed uces th e 17-
bi t d ata to a 1-bi t s tr e am by u sing EA mod u l at ion. I n th is p rocess q uant iz at ion
noise is r edis tr ib ut ed a w ay from th e au dio fr eq u e ncy by a s mu ch a s 2 or de r s of
ma g n i tu de . T h e bi t s tr e am is th e n con ve rt ed to an ana log form by a sw i tch ed
ca p acitor n e t work . A block diagram of th e P DM conve rt e r is s how n in Fig. 3.
16-BIT
44.1 KHZ
THI RD-ORDER
ANALO G
L OW-PAS S FILTER
L
ANALOG OUTP U T
R
32X OVER SAMPL ING
(L INEAR INT ERPO LAT OR )
2X OVERSAM PLIN G
(SAMPL E AND HOLD)
Figure 3. Blo ck diag ra m of a PDM digit al-to-analog c onv erter.
Becau se th e r e ar e only t wo volta ge r efe r e nces in th e P DM conve rt e r, th e r e is no
level mat ching r eq uir e m e nt for im p rove d accura cy. T h e r efor e th e li n e ar ity e rr or s
a ssociat ed wi th it ar e eli minat ed .
Com p ar ison s of T H D an d li n e ar i t y e rr or for v ar iou s 16-, 18-, 20-, an d 1-bi t
con ve rt e r s yiel d i nt e r es t i n g r es u l t s. P WM an d P DM con ve rt e r s s how < 1 dB
li n e ar ity for in p ut sig na ls from -100 to -80 dB an d ar e vi rtua lly li n e ar th e r e aft e r.
Som e of th e mos t expe n sive playe r s on th e mark e t wi th 18- an d 20-bi t conve rt e r s
u sing 4-, 8-, 16-, an d eve n 32-tim es ove r s am pli ng yiel d u p to 4 dB li n e ar ity e rr or
for sig na ls a s hig h a s -75 dB. I n th e THD tes t s pe rform ed wi th a -60 dB 1 kH z sine
w ave t es t sig na l, th e expe n sive mu lti-bi t playe r s s howed harm onics u p to th e 13 th
14
at level s gr e at e r that -110 dB
5
. O nly th e P DM conve rt e r w a s a ble to hold all non-
fun d am e nta l harm onics un de r -110 dB.
Digit a l Filt ering , Ov ers amp ling , a n d Nois e Sh ap ing
Ove r s am pli ng is not man d at ed by an y th eor e m discu ssed p r eviou sly, b ut it s u se
yiel ds tr e m e n dou s pe rforman ce g a i n s r eg ar dle ss of th e t ype of con ve rt e r u sed .
Ove r s am pli ng q uit e sim ply m e an s u sing a s am pli ng fr eq u e ncy gr e at e r than that
dictat ed by th e Nyq uis t th eor e m. By exceedi ng th e Nyq uis t fr eq u e ncy, man y of th e
p r ecision de man ds ma de by th e th eor e m can be r el axe d (li ke th e b rickw all fil t e r).
I n a dd i t ion to th e be n efi t s see n at th e out p ut fil t e r, th e sig na l-to-noise rat io is
boos t ed gr e at ly an d q uant iz at ion noise is r ed uced in th e au dio b an d. T h e latt e r is
decr e a sed by an i ncr edible am ount w h e n ove r s am pli n g is u sed i n conj un ct ion
wi th noise-s ha pe r s, w hich will be discu ssed s hort ly.
T h e ove r s am pli n g p rocess is well s u i t ed to a digi ta l sig na l p rocess or (DS P) ,
w hich esse nt ial tak es in au dio s am ple s, pe rform s an ope rat ion on th e m, an d th e n
out p ut s au dio s am ple s. Becau se th e s am ple s ar e modifie d, th e DS P is in effect a
digi ta l fil t e r. T h e DS P is be n eficial becau se th e ope rat ion s it pe rform s ar e p r ecise
an d r epe ata ble, not oth e r wise poss ible wi th ana log t echn iq u es , an d r es ult in lowe r
noise an d dis tort ion than wi th ana log t echn iq u es . T h e ove r s am pli ng p rocess can
be viewe d si m ply a s i nt e r le a vi n g ze ros be t wee n e ach s am ple wi th a dd i t iona l
s am ple s. I n p ra ct ice, th ese n ew s am ple s ar e p rod uced by u si n g a s h ift r egis t e r
(w h ich act s a s a de l a y li n e), coefficie nt mu l t iplie r s, an d an a dde r . T h e s h ift
r egis t e r ha s ta ps aft e r e ach delay ele m e nt . T h e out p ut of e ach ta p is tak e n an d
th e n mu l t iplie d by a coefficie nt s tor ed i n ROM a ss oci at ed wi th th e i m p u lse
r esp on se of th e low-p a ss fil t e r . T h ese de l a ye d mu l t iple s ar e th e n s umm ed to
ge n e rat e a n ew s am ple. An ex am ple of th is can be see n in Fig. 4.
15
T
1
one-
sample
delay
T
2
one-
sample
delay
T
3
one-
sample
delay
T
4
one-
sample
delay
S
A
M
P
L
E
V
A
L
U
E
S
A
M
P
L
E
V
A
L
U
E
S
A
M
P
L
E
V
A
L
U
E
Input
(16-bit Samples
@ 44.1 kHz)
Filter
Coefficients
C
1 X
C
2 X
C
3 X
C
4 X
+
Output
@ n44.1kHz
Figure 4. Use of a tra nsversal filter to a c hieve oversa m pling .
T h e tota l r es u l t of th is p rocess is that n ew i nt e r pol at ed s am ple s ar e cr e at ed at
e ach int e rle ave d ze ro-valu e. T his is s how n gra p hically in Fig. 5.
TIME TIME TIME
(a) (b) (c)
Figure 5. Eff e c t of zeros int erle a vin g a n d oversa m plin g on a
sig n al. Origin al sig n al a n d sa m ples (a) with: int erle a ve d zeros
(b) and int erp olate d new sa m ples (c).
As a r es ult of th is, th e s am pli ng fr eq u e ncy ha s incr e a sed by w hat eve r am ount of
ove r s am pli n g occurr ed , an d th e d ata wor d le n g th ha s g row n . Becau se th e
s am pli ng fr eq u e ncy ha s rise n, th e noise in th e au dio b an d ha s bee n s hift ed out by
a g r e at e r am ount than i t w a s befor e. Noise s ha pi n g is th e n i m ple m e nt ed to
r ed uce th e d ata wor d size an d furth e r ex agge rat e th e am ount noise is move d out
of th e au dio b an d.
As s tat ed p r eviou sly, th e p rimar y job of th e noise s ha pe r is to alt e r th e fr eq u e ncy
spe ctrum of th e e rr or sig na ls so a s to move mos t of th e q uant iz at ion e rr or out of
th e au dible fr eq u e ncy ran ge. Noise s ha ping r ed uces q uant iz at ion noise by u sing a
n eg at ive feedb ack t echn iq u e. I n effect, th e s ha pe r att e m p t s to r ed uce q uant iz at ion
e rr or by u sing it s kn ow n q ua li tie s to actua lly s u b tra ct from th e sig na l. T h e powe r
be h i n d a low-bi t con ve r sion t echn iq u e r elies on th e powe r of i t s noise-s ha pi n g
16
R
E
L
A
T
I
V
E
N
O
I
S
E
D
E
N
S
I
T
Y
R
E
L
A
T
I
V
E
N
O
I
S
E
D
E
N
S
I
T
Y
a lgor i thm . I n ge n e ra l, th e mor e com plex th e noise-s ha pe r, th e lowe r th e au dio
b an d noise. T hu s th e pe rforman ce of th e noise s ha pe r is de t e rm in ed by th e or de r
of th e s ha pe r an d it s ope rat ing fr eq u e ncy. T h e latt e r p aram e t e r is a fun ction of
how mu ch ove r s am pli ng is pe rform ed p rior to s ha ping. T h e fir s t r el at ion s hip we
can ex tra ct from th ese p aram e t e r s is th e h ig h e r th e or de r of th e s ha pe r , th e
hig h e r th e slope of th e noise r edis tr ib ut ion an d h e nce th e lowe r th e au dible noise.
T h e d ra wb ack is that sideb an d noise is incr e a sed so mu ch that th e ana log fil t e r s
could be ove r b ur de n ed . T h e secon d r el at ion s hip is that th e hig h e r th e ope rat ing
fr eq u e ncy, th e h ig h e r i n th e fr eq u e ncy doma i n th e noise is s h ift ed . T h ese t wo
r el at ion s h ips ar e defi n ed by th e noise-de n si t y dis tr ib ut ion eq uat ion w h ich is
s how n in E q. 4, w h e r e f
s
is th e origi na l s am pli ng fr eq u e ncy an d n is th e s ha pe r
o r de r ..
n
| f |
(
noise
( f ) =
2 sin
| ( (4)
\ f
s
.
T h e r el at ion s h ip is a lso ill u s trat ed i n F ig. 6 a. T h e on ly li m i tat ion i n ope rat ion
speed is th e avail a ble speed of digi ta l logic. T h e r efor e, th e con scie nt iou s des ig n e r
aim s for th e p rope r b alan ce be t wee n s ha ping or de r an d ove r s am pli ng.
PWM / MASH: THIRD-ORDER NOISE SHAPER
8 AT 32X f
s
(1.4112 MHz)
0.012
0.010
6
PDM BITSREAM:
SECOND-ORDER NOISE SHAPER
4
AT 256X f
s
(11.2896 MHz)
0.008
0.006
0.004
2
PWM / MASH: THIRD-ORDER
NOISE SHAPER AT 32X
WITHOUT NOISE SHAPER
0
0.002
0
PDM BITSREAM: SECOND-
ORDER
NOISE SHAPER AT 256X
0 2 4 6 8 10 12
FREQUENCYMHz
(a)
0 10 20 30 40 50
FREQUENCYkHz
(b)
Figure 6. Various noise-d ensity distributions as a fun c tion of fre quen cy
(a). The PDM and PWM distributions in the audio ba nd (b).
17
d
dt
}
}
T
As a foo tn o t e, th e ope rat i n g f r eq u e n cy ha s th e g r e at es t effec t of th e t wo
p aram e t e r s on noise de n sity dis tr ib ut ion. T his is cle ar ly visible in a mu ch mor e
de ta ile d look at th e noise dis tr ib ut ion s i n F ig. 6b . Cle ar ly, th e P DM ha s
sig n ificant ly lowe r au dio b an d noise an d n ecessi tat es on ly a si m ple ana log
r econ s tru ction fil t e r. A block diagram of th e th ir d-or de r noise s ha pe r u sed in th e
MAS H conve rt e r is s how n in Fig. 7.
FIRST-ORDER NOISE SHAPER
X(Z) +
+
} Q
1
INTEGRATOR
+
Y(Z)
T
DELAY
+
SECOND-ORDER NOISE SHAPER
+
+
+
+
INTEGRATOR
Q
2
INTEGRATOR
DELAY
Figure 7. Third-ord er noise sh ap er use d in th e PWM c onv erter.
I n th e s ha pe r give n in Fig. 6, th e in p ut sig na l is fed int o q uant ize r Q1 aft e r th e
r esid ua l e rr or is s u b tra ct ed from th e de l a y block i n th e fi r s t or de r s ha pe r. T h e
r esid ua l sig na l is also fed int o th e secon d or de r noise s ha pe r, w h e r e th e out p ut of
th e secon d q uant ize r, Q2, is diffe r e nt iat ed an d th e n s umm ed wi th th e out p ut of
th e fir s t noise s ha pe r to cre at e th e fina l out p ut sig na l
6
.
T h e com p act disc ha s only exis t ed for a bout 13 ye ar s, an d mor e than li kely ha s
a s man y ye ar s of u seful life left. T h e r e ar e man y a dvan ces that ar e s till poss ible in
th e fo rmat an d man y of th e m ar e j u s t i n th ei r i n f an cy. H oweve r , man y
cha lle nge r s ha ve alr e a dy e nt e r ed th e playi ng fiel d; som e by th e origi na l cr e at or s
of th e com p act disc. Sony ha s cr e at ed both th e DAT s tan d ar d a s well a s th e Mi ni-
Dis c, an d Ph ili ps ha s cr e at ed th e DCC (digi ta l com p act ca sse tt e). Reg ar dle ss of
18
th e com p act discs life tim e, it is ce rta in that digi ta l au dio will r e ma in, an d ana log
will be r ese rve d to th e role of in p ut at th e microp hon e in th e s tu dio an d out p ut at
th e spe ak e r in th e lis t e ning e nvi ronm e nt .
T his is by no m e an s a com ple t e or ex hau s tive ana lysis of th e b a sic fun d am e nta ls
of th e com p act disc pl a ye r. M an y iss u es s uch a s e rr or-corr ect ion, d ata e ncodi n g
an d decoding, an d pickup des ig n we r e n eglect ed . Howeve r, th e concep t s cove r ed
h e r e s hou ld p rovi de th e r e a de r wi th a s tr on g b ack g roun d, an d i nci t e som e
int e r es t in le arn ing mor e. For th e r e a de r w ho is int e r es t ed in le arn ing mor e, th e
Th e Art of Digital Aud io by Watk in son is an ex t e n sive collection of kn owledge on
digi ta l au dio. I t is at tim es ve ry t echn ical in natur e, b ut th e mat e rial intr od uced
b uilds u pon it self nicely. Pohlmann s book, Th e Co m pact Dis c H a nd book, focu ses
solely on th e com p act disc pl a ye r an d th e com p act disc i t se lf a lon g wi th a ll i t s
dive r se format sof w h ich au dio is on ly on e. H is book is ve r y th orou g h i n i t s
cove ra ge an d s h o u l d le a ve n o q u es t io n s f r o m th e r e a de r unan swe r ed .
Pohlmann s book ha s a fair am ount of ove rla p wi th Watk in sons an d would mak e
a be tt er s tart ing point for th ose s hort on time.
EN DNOT E S
1
John Ear gle. (Bi tt er J itt er an d Swee ter Di th er, Aud io. Vol. 76, No. 1). p. 24
2
Ke n C. Pohlmann . (Th e Co m pact Dis c H a nd bookTh e Co m p u ter m u sic a nd d igital a ud io series.
2 n d E d. M a dison, WI: A-R E dition s, I nc. 1992 ). p. 34-35.
3
Pohlmann , p. 46.
4
Pohlmann , p. 148 .
5
Pra s anna S hah . (M u sic of th e Bi t s tr e am , Aud io. Vol. 75, No. 1). p. 64.
6
Shah , p. 61.
19
REFEREN CES
Ear gle, J ohn . Bi tt e r J itt e r an d Swee t e r Di th e r, Aud io. Vol. 76, No. 1 ( J an . 1992 ).
24+
Oppe nh ei m, Al an V. S ig n als a nd S yst e m s. En glew ood Cliffs, N J : Pr e nt ice-Ha ll,
I nc., 1983 .
Poh l mann , Ke n C. Th e Co m pa ct Dis c H a nd boo k Th e Co m p u ter m u si c a nd
d igital a ud io series. 2 n d E d. M a dison, WI: A-R E dition s, I nc. 1992 .
S hah , Pra s anna . M u sic of th e Bi t s tr e am , Aud io. Vol. 75, No. 1 ( J an . 1991 ). 56 -
60+.
S tra w n , J ohn . Digital Aud io En gi neeri n g Th e Co m p u ter m u sic a nd d igital
a ud io series. Los Altos, CA: Willi am K au fmann , I nc., 1985 .
S tra w n, J ohn . Digital Aud io S ig n al Processi ngTh e Co m p u ter m u sic a nd d igital
a ud io series. Los Altos, CA: Willi am K au fmann , I nc., 1985 .
Watk in son, J ohn R. Th e Art of Digital Aud io. 2 n d E d.. Bos ton, MA: Focal Pr ess ,
1994.
Watk in son, J ohn R. Cod i ng for Digital Recor d i ng. Bos ton, MA: Focal Pr ess , 1990 .
Whyte, Be rt . Eve ry Li tt le Bi t Helps , Aud io. Vol. 76, No. 8 (Aug. 1992 ). 19-21.