LCD Music Display and Graphic Equalizer

1
LCD Music Display and Graphic Equalizer

• Adam Krzesinski
• andCristina Vasco

2
Introduction

• Purpose
• Graphical display of frequency components present
  in audio signal
• Visual representation of audio levels to aid in
  setting EQ parameters
• Real time equalization of audio signal
• EQ frequency bands correspond to displayed bands

3
Features LCD Display

• HD44780 display controller
• 4 channels monitored
• Amplitude resolution - 16 levels
• Highly responsive with imperceptible delay (lt
  30ms)

4
Features Equalizer

• Four modification bands that correlate to display
• FFT Convolution method equivalent to 500-tap FIR
  filter
• User input through faders

5
Filterbank Original Design

• 8 hardware oscillator filters LCR filter design
• Output reflects energy of input signal in
  different frequency ranges
• Constant Q over 8 octaves
• 1 lowpass, 1 highpass, 6 bandpass

6
Filterbank - Problems

• Impractical L and C values for low-frequency
  filters eg. 4.4mH
• Large component values would produce unresponsive
  display
• Large physical size of inductors and capacitors
  leading to expensive or inaccurate component
  values
• 4.4mH gt 878 turns with largest AL available
  (core T68-2)

7
Filterbank DSP Implementation

• Reduced to 4 bands
• Operates on averaging FFT energy in each band
• Each channel corresponds to trace of FFT in one
  of four frequency ranges
• PIC interprets instantaneous DC values for
  display as magnitude of bar

8
Filterbank DSP Implementation

• Response of 4 filters to log chirp stimulus

9
Filterbank DSP Implementation

• Frequency bands chosen to cover equal space in
  log-frequency
• Each band should also cover ranges with similar
  human hearing sensitivity

10
Filterbank DSP Implementation

• FFT energy ? loudness
• Capture above shows most energy is focused in low
  frequencies
• Display should parallel perceptual loudness of
  each frequency band
• Solution amplify higher bands before display

11
DSP Equalizer FFT Convolution

• Manipulation of FFT samples frequency domain
• Equivalent to convolution time domain operation
• Speed advantage when implementing higher-order
  filters
• ( gt 60 taps)
• Algorithm explanation at http//www.dspguide.com/C
  H18.PDF

12
DSP Equalizer FFT Convolution

• Figure 18-2 from http//www.dspguide.com/CH18.PDF
  - The Scientist and Engineer's Guide to Digital
  Signal Processing by Steven W. Smith, Ph.D.

13
Challenges

• AC coupling on DSP
• Since signal is fluctuating, average is zeroed
  out by DSPs AC coupling
• Should just be interpreted as non-negative DC
  value
• Result is negative when signal should small or 0
  Volts
• Solution - change negative reference voltage on
  PIC

14
Challenges AC Coupling on DSP

• Cursors show LCD driver output is -359mV at times
• -333mV reference used for PIC to ensure negative
  values are displayed as non-negative

15
Reference Voltage Circuits for Display PIC
16
Challenges

• LCD Speed
• Original PIC program was written to accommodate
  worst case timing for LCD display (anticipated
  10ms draw time)
• Sluggish performance with original code
• PIC program had to be optimized to exit display
  interrupt as quickly as possible
• Solution is to poll busy flag generated by LCD
• Result is high frame rates and overall
  responsiveness of LCD

17
Challenges

• DSP Fixed-point arithmetic
• FFT scales outputs by 0.5 after each stage to
  avoid overflow
• Radix-2 1024-pt FFT gt 10 stages gt loss of 10
  bits of precision of 16 bits possible
• Magnitude scaled by 1/20
• Output is barely audible and pervaded by
  quantization noise, evident after amplification
• FFT Convolution algorithm works well in Matlab
  where environment is floating-point

18
Challenges

• Graphic LCD
• Higher resolution and more flexible for
  applications with complex graphics
• Insufficient documentation
• Manufacturers site lacked a quality datasheet
• Solution
• HD44780 character display
• Very well documented with many examples
• Suitable for our application

19
Challenges

• Serial Communication
• Did not fully implement PIC to DSP serial
  communication
• PIC was able to generate a stream of serial bits
  but DSP code incorrectly interpreted them

20
Successes

• Equal log-spaced filters as desired
• Filters cover entire audible range
• Responsive LCD Display
• High quality equalizer results in Matlab
• PIC able to output information serially

21
Recommendations

• Implement both display and EQ with embedded
  system
• More channels of display and control
• Larger memory capacity than DSP board for more
  accurate FFTs, more channel buffers
• Floating-point CPU for preserving information
  when manipulating audio through FFT

22
Thank You

• Questions?