7.3 Digital Signal Processing & Compression Architectures

7.3 Digital Signal Processing & Compression Architectures

[!NOTE] A damaged cochlea does not simply act as an attenuated microphone; it suffers from a fundamental loss of dynamic range termed recruitment. In the normal ear, the auditory system processes acoustic intensities spanning from 0 dB SPL (threshold of audibility) to 100-120 dB SPL (threshold of discomfort)—a dynamic range of over 100 dB. In sensorineural hearing loss, outer hair cell loss elevates hearing thresholds while upper loudness tolerance limits remain largely unchanged. Digital Signal Processing (DSP) compression architectures are engineered to compress this wide environmental dynamic range into the patient's shrunken residual auditory window without creating harmonic distortion.

---

Linear Amplification vs. Non-Linear Compression

The Mechanics of Linear Amplification

In a linear hearing instrument, the relationship between acoustic input and acoustic output is strictly constant. The amplifier applies the exact same decibel gain regardless of whether the incoming sound is a whisper (40 dB SPL), conversational speech (65 dB SPL), or a slamming door (90 dB SPL):

$\text{Output (dB SPL)} = \text{Input (dB SPL)} + \text{Gain (dB)}$

$\text{Gain} = \text{constant} \quad (\Delta \text{Input} = \Delta \text{Output})$

On an Input-Output (I/O) graph, linear amplification is represented as a straight diagonal line with a 1:1 slope ($45^\circ$ angle). If an instrument provides 30 dB of linear gain:

• A 40 dB SPL input yields a 70 dB SPL output.
• A 60 dB SPL input yields a 90 dB SPL output.
• An 80 dB SPL input yields a 110 dB SPL output.

                  LINEAR VS. COMPRESSION INPUT-OUTPUT FUNCTIONS

        Output
        (dB SPL)
         130 ┼───────────────────────────────── Peak Clipping / Saturation
             │                                ╭──────── Compression Limiting
         110 ┼──────────────────────────────╭─╯
             │                            ╭─╯
          90 ┼──────────────────────────╭─╯ ◄── Compression Threshold (TK)
             │                        ╭─╯       (Slope < 1:1, e.g. 2:1)
          70 ┼──────────────────────╭─╯
             │                    ╭─╯
          50 ┼──────────────────╭─╯ ◄── Linear Region (Slope = 1:1)
             │                ╭─╯
          30 ┼──────────────╭─╯
             └──────────────┼───────────┼───────────┼───────────┼────
                           30          50          70          90   Input (dB SPL)

Why Linear Amplification Fails Sensorineural Loss

Linear amplification fails sensorineural pathology due to cochlear recruitment:

1. If gain is set high enough to make faint, high-frequency speech phonemes (unvoiced fricatives like /s/, /f/, /θ/ at 40 dB SPL) audible, normal conversational speech (65 dB SPL) becomes uncomfortably loud, and shouting or traffic (85-90 dB SPL) instantly reaches intolerable, painful levels.
2. If gain is reduced so that loud sounds remain tolerable, faint speech sounds drop below the patient's elevated hearing threshold and become completely inaudible.
3. The patient is trapped in a cycle of constant manual volume wheel manipulation.

Peak Clipping and Harmonic Distortion

When a linear hearing aid attempts to deliver an output that exceeds the battery voltage rails or the receiver's physical magnetic limits, the amplifier undergoes peak clipping. The tops and bottoms of the acoustic sound wave are abruptly cut off into flat, squared-off plateaus:

• Odd Harmonic Generation: A clipped sine wave morphs into a square wave, generating intense energy at odd harmonic frequencies ($3^{\text{rd}}, 5^{\text{th}}, 7^{\text{th}}$ multiples of the fundamental).
• Intermodulation Distortion (IMD): When complex multi-tone signals like speech or music are clipped, the frequencies interact to generate phantom difference tones ($f_2 - f_1, 2f_1 - f_2$).
• Peak clipping results in a harsh, scratchy, metallic timbre that obliterates speech recognition and causes extreme listening fatigue.

---

Wide Dynamic Range Compression (WDRC)

The Physiological Objective: Managing Recruitment

In a healthy cochlea, approximately 12,000 to 15,000 outer hair cells (OHCs) act as active biological electromotors (the "cochlear amplifier"). They sharply amplify faint sounds (providing up to 40-50 dB of biological gain) while providing zero amplification to loud sounds. This active basilar membrane mobility is inherently compressive and non-linear.

When sensorineural hearing loss destroys outer hair cells, this active non-linear amplification is lost. The ear loses both acoustic sensitivity and dynamic range. Wide Dynamic Range Compression (WDRC) is an electronic emulation of the missing outer hair cells:

• Faint Inputs (35-50 dB SPL): Provided with maximum acoustic gain to guarantee that low-intensity consonant cues cross the patient's elevated hearing threshold.
• Moderate Inputs (60-70 dB SPL): Provided with moderate gain to place conversational speech at the patient's Most Comfortable Loudness (MCL) level.
• Loud Inputs (80-95 dB SPL): Provided with minimal or unity gain (0 dB), preventing sounds from exceeding the patient's Loudness Discomfort Level (LDL).

┌─────────────────────────────────────────────────────────────────────────────┐
│                     WDRC COMPRESSIVE GAIN BEHAVIOR                          │
├──────────────────┬──────────────────┬──────────────────┬────────────────────┤
│ Input Sound      │ Acoustic Level   │ Hearing Aid Gain │ Resulting Output   │
├──────────────────┼──────────────────┼──────────────────┼────────────────────┤
│ Soft Whisper     │ 45 dB SPL        │ 35 dB Gain       │ 80 dB SPL (Audible)│
│ Average Speech   │ 65 dB SPL        │ 25 dB Gain       │ 90 dB SPL (Comfort)│
│ Shouting / Horn  │ 85 dB SPL        │ 10 dB Gain       │ 95 dB SPL (Safe)   │
└──────────────────┴──────────────────┴──────────────────┴────────────────────┘

Note: As acoustic input increases from 45 to 85 dB SPL (a 40 dB increase), the output only increases from 80 to 95 dB SPL (a 15 dB increase). The wide 40 dB acoustic input range has been compressed into a narrow 15 dB auditory window.

---

Compression Parameters: TK, CR, and Output Limiting

                        ANATOMY OF A COMPRESSION CURVE
       Output (dB SPL)
        120 ┼────────────────────────────────── Output Limiting (CR = 10:1)
            │                              ╭───
        100 ┼────────────────────────────╭─╯
            │                        ╭───╯ ◄── WDRC Region (CR = 2:1)
         80 ┼──────────────────────╭─╯
            │                  ╭───╯ ◄── Compression Kneepoint (TK = 45 dB SPL)
         60 ┼────────────────╭─╯
            │              ╭─╯ ◄── Linear Gain Region (CR = 1:1)
         40 ┼────────────╭─╯
            │          ╭─╯ ◄── Expansion Kneepoint (TK_exp = 35 dB SPL)
         20 ┼────────╭─╯
            └────────┼─────────┼─────────┼─────────┼─────────┼────
                    20        40        60        80       100   Input (dB SPL)

1. Compression Threshold / Kneepoint (TK)

The Compression Threshold (TK)—also called the kneepoint—is the sound pressure level on the input-output function where the amplifier transitions from linear amplification to compressive amplification:

• Low Kneepoint (TK = 40 to 50 dB SPL): The foundational hallmark of WDRC. By placing the kneepoint at or below the level of faint conversational speech, nearly the entire speech spectrum receives compressive processing. Soft consonants receive full amplification, restoring audibility without manual intervention.
• High Kneepoint (TK > 65 to 70 dB SPL): Characteristic of compression limiting (OCL). The hearing aid behaves linearly for soft and conversational speech, maintaining natural speech dynamics. Compression engages exclusively when input levels threaten to exceed comfort limits.
• Expansion (Low-Level Squelch): Applied to very faint inputs below 30-35 dB SPL. Expansion applies a compression ratio of less than 1:1 (e.g., 1:2), meaning gain is actively reduced as inputs drop. This squelches internal microphone circuit hiss and annoying low-level ambient environmental hums (e.g., refrigerator compressors, distant HVAC units) that normal-hearing listeners ignore but hearing aid users find distracting.

2. Compression Ratio (CR)

The Compression Ratio (CR) defines the mathematical relationship between a change in input sound pressure level and the resulting change in output sound pressure level above the compression kneepoint:

$\text{CR} = \frac{\Delta \text{Input (dB)}}{\Delta \text{Output (dB)}}$

Quantitative Clinical Derivations

• Case Example 1: An input level increases by 20 dB (from 50 dB SPL to 70 dB SPL). The measured hearing aid output increases by only 10 dB (from 75 dB SPL to 85 dB SPL): $\text{CR} = \frac{70 - 50}{85 - 75} = \frac{20\text{ dB}}{10\text{ dB}} = 2.0:1$
• Case Example 2: An input increases by 30 dB (from 60 dB SPL to 90 dB SPL), and the output increases by 5 dB (from 95 dB SPL to 100 dB SPL): $\text{CR} = \frac{90 - 60}{100 - 95} = \frac{30\text{ dB}}{5\text{ dB}} = 6.0:1$

Clinical Compression Ratio Guidelines

• Low Compression Ratios (1.2:1 to 2.5:1): Standard for WDRC across multichannel speech frequencies. Ratios under 2.5:1 preserve the natural temporal envelope, vowel-to-consonant acoustic contrast, and amplitude modulation of speech. Excessive compression ratios (>3:1) in WDRC flatten the speech envelope, smearing acoustic contrast and degrading word recognition in background noise.
• High Compression Ratios (>5:1 to 10:1 or $\infty:1$): Used exclusively in Output Compression Limiting (OCL) to cap maximum output below the patient's Loudness Discomfort Level (LDL).

---

Compression Dynamic Time Constants: Attack & Release Times

Compression does not occur instantaneously; it is governed by temporal filters designated as attack time and release time, standardized under ANSI S3.22 electroacoustic measurement protocols.

                       ANSI S3.22 ATTACK & RELEASE TIMES

     Input Step:       [55 dB SPL] ──► [90 dB SPL Transient] ──► [55 dB SPL]

     Output Waveform:
     ───┐              ╭─────────────────╮                      ┌────────
        │             ╭╯ ◄── Overshoot   ╰╮                     │
        │            ╭╯                   ╰╮                    │
        └────────────╯                     ╰────────────────────┘
                     ▲                    ▲
                     │◄── Attack Time ──►││◄─── Release Time ───►│
                     │    (Settles to     │    (Recovers to       │
                     │    within 3 dB)    │    within 4 dB)       │

Definitions Under ANSI S3.22

1. Attack Time ($t_a$): The time required for the hearing aid amplifier to reduce its gain to within 3 dB (or 2 dB) of its final steady-state compressed output following an instantaneous increase in input level from 55 dB SPL to 90 dB SPL.
   • Modern digital hearing aids feature attack times of 1 to 10 milliseconds (ms).
   • A fast attack time is clinically non-negotiable: if an attack time is too slow (>20-30 ms), a sudden acoustic transient (e.g., a dropped ceramic plate, a gunshot, a car horn) will be amplified with full linear gain before the compressor reacts, causing acute loudness pain and risking acoustic trauma.
2. Release Time ($t_r$): The time required for the amplifier to restore its gain to within 4 dB of its final steady-state uncompressed output following an instantaneous drop in input level from 90 dB SPL back to 55 dB SPL.
   • Release times vary widely based on circuit design, ranging from 20 ms to 3000 ms (3 seconds).

Syllabic vs. Slow Dual/Adaptive Time Constants

The Phenomenon of Compression "Pumping" (Breathing)

When a fast syllabic release time ($t_r < 50\text{ ms}$) is employed, the compressor rapidly boosts gain during the brief silent pauses between spoken words. In a restaurant or room with steady background ventilation, the listener hears the ambient background noise surge upward ("whoosh") during every conversational pause, only to be abruptly driven downward when the speaker articulates the next vowel. This distracting auditory artifact is known as compression pumping or breathing.

Modern digital hearing aids eliminate pumping by utilizing dual or adaptive time constants: a short acoustic transient (like a clattering dish, duration <50 ms) triggers an immediate fast release (30 ms) so conversational speech is not muted, whereas continuous conversational speech engages a slow release time (1.5 seconds) to keep gain stable.

---

Output Compression Limiting (OCL) vs. Peak Clipping

Protecting the patient from exceeding their Loudness Discomfort Level (LDL / UCL) is the primary safety directive in hearing aid programming. Two distinct mechanisms achieve this:

                     PEAK CLIPPING VS. COMPRESSION LIMITING

           PEAK CLIPPING                          OUTPUT COMPRESSION LIMITING
        (Waveform Distortion)                           (Clean Output)

          /\        /\                              /\        /\      
       ──/──\──────/──\── Capped Rail            ──(  )──────(  )── Compression
        /    \    /    \                          /    \    /    \  Ceiling (Clean)
       /  ──  \  /  ──  \                        /      \  /      \   
       ───────────────────                       ───────────────────  
       [Square-wave clipping:                    [Waveform preserved: 
        High THD (>20%), raspy]                   Low THD (<1%), natural]

Input Compression (AGC-I) vs. Output Compression (AGC-O)

• Input Compression (AGC-I): The level detector is placed before the volume control. The compression threshold is tied directly to incoming environmental sound pressure. Adjusting the user volume control shifts both conversational speech and maximum output together, but does not alter the compression kneepoint.
• Output Compression (AGC-O / OCL): The level detector is placed after the volume control and power amplifier, directly monitoring the sound pressure level delivered to the receiver. The kneepoint is calibrated strictly relative to maximum acoustic output (OSPL90).

Why OCL is Superior to Peak Clipping

In Output Compression Limiting (OCL), when the output signal approaches the patient's measured LDL, the detector engages an ultra-high compression ratio (e.g., $10:1$ to $\infty:1$). Rather than mechanically cutting off the waveform peaks, the DSP instantaneously turns down the pre-amplifier gain.

• The entire acoustic waveform is scaled downward in amplitude, preserving the sinusoidal integrity of the signal.
• Total Harmonic Distortion remains under 1% to 2%, delivering clean, transparent, distortion-free sound while providing a rigid, impenetrable acoustic ceiling that ensures loud sounds never cross the threshold of discomfort.