Convention E-Brief 631 Presented at the 149Th Convention 2020 October 21 – 24, New York, NY This Engineering Brief Was Selected on the Basis of a Submitted Synopsis

Convention E-Brief 631 Presented at the 149Th Convention 2020 October 21 – 24, New York, NY This Engineering Brief Was Selected on the Basis of a Submitted Synopsis

Audio Engineering Society Convention e-Brief 631 Presented at the 149th Convention 2020 October 21 – 24, New York, NY This Engineering Brief was selected on the basis of a submitted synopsis. The author is solely responsible for its presentation, and the AES takes no responsibility for the contents. All rights reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the Audio Engineering Society. Generalized Digital Second Order Systems Beyond Nyquist Frequency Champ Darabundit1,2 and Russell Wedelich2 1CCRMA, Stanford University 2Eventide Inc. Correspondence should be addressed to Champ Darabundit ([email protected]) ABSTRACT The Bilinear transform is an established design method for producing digital representations of analog filters. But due to frequency warping inherent in the transform, adequate representation in the digital domain is limited for analog filters designed near or above the Nyquist frequency. Previous work has demonstrated a method for surpassing this design limitation in the case of parametric filters. This paper is an extension of that previous work, and proposes a design methodology for digitally representing any second order system with a center frequency near or above the Nyquist frequency. By repositioning the system poles, this method compensates for the effects of frequency warping, and allows for a more exact digital replication of an analog system’s magnitude response. 1 Introduction systems and correct for the frequency warping inherent in the BLT. The bilinear transform (BLT) is a prevalent technique The motivation for this work was the modeling filters for the digitization of analog filters. The BLT maps prior to an analog-to-digital converter (ADC) and fol- the highest analog frequency (s = ¥) to the highest lowing a digital-to-analog converter (DAC). The sys- digital frequency (z = −1) and the analog DC (s = 0) to tem being modeled has a sampling rate of 44.1 kHz digital DC (z = 1), and avoids any aliasing that occurs and uses filters to condition the signal for the ADC in normal sampling [1]. A limitation of the BLT is that and DAC around the system’s Nyquist frequency of mapping from an infinite analog frequency space to a 22.05 kHz. While modeling these filters and accurately periodic discrete frequency space results in frequency digitizing their responses is possible at higher sampling warping that is especially pronounced near the Nyquist frequencies, due to frequency warping, the analog filter frequency. Sierra [2] proposes a method to correct for behavior was not accurately replicated at 44.1 kHz and the effects of frequency warping in parametric filters 48 kHz. A direct solution would be upsampling the with center frequencies above the Nyquist frequency. signal processing for these operating frequencies, but This paper proposes a general method for digitizing any this was deemed undesirable due to increased computa- second-order systems with analog center frequencies tional complexity and the inelegance of implementing near or above the Nyquist frequency through reposi- separate algorithms for different sampling frequencies. tioning the system poles. This allows us to make use of By utilizing the proposed method, the same implemen- the damping ratio which characterizes all second-order tation for a filter can be used across different sampling Darabundit and Wedelich Digital systems beyond Nyquist frequency frequencies, and the analog behavior is well replicated frequency, Sierra proposed the following definition for across all implementations. the Bilinear constant: −( w )2 Section 2 will cover prior work, reviewing the BLT and c = 2 fse p fs (6) Sierra’s methodology. In Section 3, we will demon- strate how Sierra’s method can be generalized for any Sierra’s Bilinear constant is stable for frequencies > second-order system. Section 4 will provide a design fs=2 as it is always greater than 0. example for digitizing a pre-emphasis filter. Section 5 Due to frequency warping inherent in the BLT, the gain offers conclusions and suggestions for future work. at the digital center frequency, wd, will differ signif- icantly from the gain at the parametric filter’s center 2 Prior Work w frequency when 2p approaches fs=2. This effect can be compensated for by scaling Q by a correction factor The methodology outlined here is used by Sierra to denoted as r. match the behavior of digital parametric filters with center frequencies near and beyond the Nyquist fre- The location of the digital center frequency is found quency to their analog responses. A review of the BLT using Sierra’s Bilinear constant. and Sierra’s work is given. −1 wd = 2 fstan (w=c) (7) 2.1 Bilinear Transform Then the digital center frequency gain is found by solv- ing Eq. 5 at w and taking the magnitude of the result. The BLT can be viewed as a substitution from the d Laplace domain to the z-domain. gd = jjH( jwd)jj (8) 1 − z−1 s = c (1) The correction factor r is dependent on the ratio of 1 + z−1 Sierra’s Bilinear constant and the sampling frequency, and a ratio of the digital center frequency gain g and c − s d z−1 = (2) the parametric filter gain g. c + s ( c g Where c is the Bilinear constant. · g > gd r = 2 fs gd (9) c gd 2 · g < gd c = (3) 2 fs g T The two conditions account for maintaining boost-cut T is the inverse of the digital sampling frequency. If a symmetry in the parametric filters. The corrected qual- perfect mapping from a singular analog frequency to a ity factor Q0 is the parametric parameter Q scaled by digital frequency is desired, then c is modified to such, r. Q0 = rQ (10) w c = (4) 0 tan(wT=2) Q along with gd are substituted into Eq. 5 to derive the corrected analog parametric filter transfer function in a technique known as frequency pre-warping. 2 w 2 s + gd 0 s + w H (s) = Q (11) 2.2 Sierra’s Method corr 2 w 2 s + Q0 s + w Sierra deals with parametric filters of the form: And this transfer function can then be digitized using the BLT with Sierra’s Bilinear constant. 2 w 2 s + g Q s + w H(s) = 2 w 2 (5) s + Q s + w 3 Proposed Generalized Method where g is the parametric filter gain, w is the center Sierra’s method adjusts the Q of the parametric filter frequency, and Q is the quality factor. To guarantee to compensate for the frequency warping at the para- stability for the Bilinear transform above the Nyquist metric filter’s center frequency. We can realize that a AES 149th Convention, New York, NY, 2020 October 21 – 24 Page 2 of 5 Darabundit and Wedelich Digital systems beyond Nyquist frequency parametric filter is a specific case of a second-order sys- wd, is dependent on our analog center frequency, wa, tem where the quality factor and center frequency are instead of the parametric filter parameter w. directly accessible in the system’s formulation. Sierra’s −1 compensation should be possible for any second-order wd = 2 fstan (wa) (18) system if Q is adjusted through direct manipulation of the system’s poles. Our correction factor r is now dependent on the sys- tem’s analog center frequency gain, ga, instead of the g 3.1 Generalization parametric filter gain parameter . ga = jjH( jwa)jj (19) The linear constant-coefficient differential equation for a second-order system is a general equation for any And, in the general case we no longer worry about second order system [3]. boost-cut symmetry. Our correction factor is thus: 2 2 y¨(t) + 2 y˙(t) + y(t) = x(t) (12) c gd zwn wn wn r = · (20) 2 fs ga The frequency response of this second order system in the Laplace domain is then: As our correction factor is only dependent on values related to the position of the system’s poles, we view 2 wn the act of correcting a second-order system’s response H(s) = 2 2 (13) s + 2zwns + wn as an adjustment of the system’s pole positions. It is useful then, to convert our transfer function model And is characterized by the undamped natural fre- to the zero-pole-gain model, shown in Eq. 21, when quency, wn, and the damping ratio, z. wn is in most correcting our analog filter. While Eq. 21 has two cases approximately equal to the generalized analog zeros, our second-order system can have one to two center frequency, wa, which is used here to draw dis- zeros but will always has two poles. Here z1;2 are the tinction from the parametric filter parameter w. The zeros and k the scalar gain of our system. damping ratio is related to Q by: 1 (s − z1)(s − z2) z = (14) H(s) = k (21) 2Q (s − p1)(s − p2) Using Eq. 14 and the quadratic equation we can derive 3.2 True peak-gain matching our poles p1;2 in terms of Q and wn. 0 s 1 For some systems, our previous approximation w ≈ 1 1 2 n p1;2 = wn @− ± − 1A (15) wa does not hold and the true peak-gain of the filter is 2Q 2Q not located at the natural frequency, and this is espe- cially true in systems where the poles are located near We can also derive the backwards relation for wn and each other [4]. It is necessary for our correction factor Q in terms of our poles. to reference the true peak-gain to properly account for p the effects of frequency warping.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    5 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us