24-Bit ADC Dynamic Range: Why You Don't Get 144 dB

The Bit Depth Marketing Problem

Every audio interface manufacturer loves to slap "24-bit" on the box. Some even claim 32-bit float these days. And sure, the math says 24 bits should give you 144 dB of dynamic range. But hook up your fancy interface to an Audio Precision analyzer, and you'll measure something closer to 110-120 dB if you're lucky.

What gives?

The gap between theoretical and measured dynamic range trips up a lot of engineers, especially those crossing over from digital signal processing into actual hardware design. The calculator we're discussing here — open the ADC Bit Depth to Dynamic Range — helps you understand what you're actually working with, and more importantly, why.

The Theoretical Foundation

Let's start with the textbook formula that everyone quotes. For an ideal ADC with uniformly distributed quantization noise, the signal-to-noise ratio works out to:

where $N$ is the number of bits. This comes from treating quantization error as uniformly distributed noise with amplitude $\\pm \frac{1}{2}$ LSB. The RMS value of that noise is $\frac{q}{\\sqrt{12}}$ where $q$ is the quantization step size, and when you compare that to a full-scale sine wave, you get the formula above.

For common bit depths:

• 16-bit: $6.02 \times 16 + 1.76 = 98.1$ dB
• 24-bit: $6.02 \times 24 + 1.76 = 146.2$ dB
• 32-bit integer: $6.02 \times 32 + 1.76 = 194.4$ dB

That 32-bit number should immediately make you suspicious. 194 dB is absurd — it's below the thermal noise floor of any real circuit by a massive margin. A resistor at room temperature generates about -174 dBm/Hz of Johnson noise. You physically cannot achieve 194 dB of dynamic range in any analog system.

Where Oversampling Actually Helps

Here's where things get interesting. Oversampling doesn't magically create more resolution, but it does let you trade bandwidth for noise performance. When you sample at a rate higher than Nyquist, the quantization noise spreads across a wider frequency band. If you then filter down to your actual signal bandwidth, you're throwing away some of that noise.

The gain from oversampling is:

where OSR is the oversampling ratio. So 4× oversampling gives you 10·log10(4) = 6.02 dB of improvement — equivalent to one extra bit. 64× oversampling (common in sigma-delta converters) gives you 10·log10(64) = 18.06 dB, or about 3 extra effective bits.

But wait, sigma-delta converters do even better than this because of noise shaping. A well-designed sigma-delta ADC pushes quantization noise up to higher frequencies where it gets filtered out, giving you significantly more than the simple oversampling gain would suggest. The calculator handles the basic oversampling math, but real sigma-delta performance depends heavily on the modulator order and loop filter design.

Worked Example: A Typical Audio Interface

Let's walk through a concrete scenario. You're designing an audio interface with a 24-bit sigma-delta ADC running at 64× oversampling (so 3.072 MHz sample rate for 48 kHz audio output).

Step 1: Calculate theoretical SNR from bit depth Step 2: Calculate oversampling gain Step 3: Total theoretical SNR

So on paper, you've got 164 dB of dynamic range. Fantastic, right?

Except no. Your analog front-end — the op-amps, the voltage reference, the anti-aliasing filter — all contribute their own noise. A typical low-noise op-amp like the OPA1612 has input noise around 1.1 nV/√Hz. Your voltage reference has noise. Your PCB layout picks up interference. The ADC's own analog circuitry adds noise before quantization even happens.

In practice, a really good audio ADC implementation might achieve 120-124 dB of dynamic range. The AKM AK5397 datasheet, for instance, claims 123 dB typical. That's excellent, but it's a far cry from 164 dB.

The calculator gives you the quantization-limited ceiling. Everything else in your signal chain will pull you down from there.

The 32-Bit Float Situation

A quick note on 32-bit float, since it shows up in the calculator options and causes confusion. 32-bit floating point isn't really comparable to 32-bit integer in terms of dynamic range.

Floating point has a 23-bit mantissa (plus an implicit leading 1, so effectively 24 bits of precision) and an 8-bit exponent. The dynamic range is enormous — around 1528 dB if you count the full exponent range — but the instantaneous SNR at any given signal level is limited by those 24 mantissa bits.

In audio applications, 32-bit float is mostly useful as a processing format to avoid clipping during intermediate calculations. No ADC actually outputs 32-bit float directly; it's a conversion done in software after the fact. When you see "32-bit float recording" in a DAW, the actual conversion happened at 24 bits (or sometimes less), and the float format just gives you headroom for processing.

Common Mistakes and Gotchas

Ignoring the Analog Front-End

This is the big one. Engineers calculate their theoretical SNR, see a beautiful number like 146 dB, and then wonder why their measured performance is 30 dB worse. The quantization noise floor only matters if it's the dominant noise source. In almost every real system, it isn't.

Before you even think about ADC bit depth, calculate your analog noise budget. What's the noise contribution from your input amplifier? Your reference? Your anti-aliasing filter? Add them up (in power, not voltage — RSS for uncorrelated sources). If that sum is higher than your quantization noise floor, more bits won't help you.

Confusing Dynamic Range with SNR

These terms get used interchangeably, but they're subtly different. SNR typically refers to the ratio of a full-scale signal to the noise floor. Dynamic range is the ratio of the largest undistorted signal to the smallest detectable signal. In an ideal ADC they're the same, but in real converters with harmonic distortion, dynamic range might be limited by THD at high signal levels rather than noise at low levels.

The calculator outputs both, but be aware that datasheets sometimes play games with which spec they're quoting.

Assuming Oversampling is Free

Oversampling requires a higher sample rate clock, more digital processing power, and a decimation filter that actually achieves the required stopband rejection. If your decimation filter only has 80 dB of stopband attenuation, you're not getting the full benefit of your oversampling ratio.

Also, oversampling helps with quantization noise but does nothing for analog noise. If your op-amp is contributing -100 dBFS of noise, oversampling by 64× won't change that number one bit.

Trusting Datasheet Dynamic Range Specs

Datasheet specs are measured under ideal conditions with carefully optimized evaluation boards. Your PCB, with its switching power supplies nearby and less-than-perfect grounding, will do worse. Budget 3-6 dB of margin between datasheet specs and what you'll actually achieve, more if your layout is compromised.

Forgetting About Dither

At low signal levels, quantization error isn't really noise — it's correlated distortion. Adding a small amount of dither (typically around 1 LSB of noise) converts this distortion into true noise, which sounds better and behaves more predictably. But that dither raises your noise floor slightly. If you're counting on every last dB of theoretical dynamic range, you need to account for this.

When More Bits Actually Matter

So if the analog front-end usually dominates, why bother with 24-bit converters at all?

Headroom. A 24-bit converter gives you about 48 dB more range than 16-bit. Even if your actual noise floor is only 110 dB down from full scale, having that extra headroom means you don't have to set your levels perfectly. You can record conservatively, leaving 20 dB of headroom, and still have plenty of resolution in your signal.

This matters a lot in live recording situations where you can't predict the exact signal level in advance. It matters less in controlled measurement applications where you can optimize gain staging.

Try It

Plug your ADC specs into the ADC Bit Depth to Dynamic Range calculator to see where quantization noise sits relative to your design targets. Then go calculate your analog noise budget and see which one actually limits your system. Nine times out of ten, it's not the ADC.