Oscilloscope bandwidth, sample rate and memory depth explained
Illustrative image: Unsplash contributor on Pexels
Oscilloscopes are often compared by three prominent numbers: bandwidth, sample rate and record length. None of them is meaningful in isolation.
Bandwidth describes the analogue measurement path. Sample rate determines how frequently the converted waveform is represented. Record length controls how many samples can be retained. The selected time span connects all three because a scope cannot maintain an unlimited sample rate across an unlimited acquisition time.
The probe, input loading, noise, triggering and waveform update rate can then determine whether the instrument captures the event the designer actually cares about.
Bandwidth is an analogue limit
Oscilloscope bandwidth is normally defined at the frequency where a sine-wave input is displayed at about 70.7% of its true amplitude, corresponding to a 3 dB reduction in power.
That means a 100 MHz scope does not reproduce a 100 MHz sine wave at full amplitude. More importantly, a digital signal is not composed only of its clock frequency. Its edges contain higher-frequency components.
When rise time matters, choose bandwidth from the edge speed rather than the repetition rate. A common first-order relationship for a Gaussian response is:
bandwidth × rise time ≈ 0.35
This describes the instrument's own response, not an assurance of measurement accuracy. To measure a signal rise time without excessive distortion, the oscilloscope should be substantially faster than the signal.
If scope and signal rise times can both be approximated as Gaussian, the displayed result is often estimated by:
measured rise time² ≈ signal rise time² + scope rise time²
The formula is useful for planning, but modern front ends and signals do not always behave as simple Gaussian systems. Manufacturer guidance should be used for critical work.
Why the usual bandwidth rule is only a starting point
Rules such as choosing three or five times the fundamental frequency can work for particular waveform shapes and accuracy goals. They fail when the edge rate is much faster than the nominal clock or when small high-frequency detail carries the fault.
A 10 MHz square wave generated by fast logic may require far more than 50 MHz of useful measurement bandwidth. Conversely, a slow sensor waveform may not benefit from a very wide front end and can show more noise when the full bandwidth is enabled.
Many scopes provide bandwidth limiting. Using it deliberately can improve the signal-to-noise ratio when higher-frequency content is not relevant.
Sample rate is not bandwidth
The sample rate states how many points the analogue-to-digital converter records per second. Nyquist theory establishes a minimum of more than twice the highest frequency for reconstructing an ideal band-limited signal. Practical oscilloscope measurements generally need more margin.
Tektronix notes that sampling at only twice the highest frequency component is usually insufficient for capturing real waveform detail. Its guidance commonly uses substantially higher oversampling for time-domain work.
The required ratio depends on the interpolation method, waveform, measurement and accuracy target. A smooth repetitive sine wave and a single fast glitch are very different problems.
Check whether the quoted maximum sample rate applies:
- with one channel active or all channels;
- in normal, high-resolution or segmented mode;
- across the chosen time base;
- to real-time acquisition or an equivalent-time mode intended for repetitive signals.
Shared acquisition hardware can reduce the per-channel rate when more inputs are enabled.
Record length decides how long full detail can be retained
Record length is the number of samples stored in one acquisition. The basic relationship is:
capture time = record length ÷ sample rate
At 1 GS/s, a one-million-point record contains one millisecond of data. Capturing one second at the same rate requires one billion points, before considering processing and display constraints.
When the time span is widened, the instrument must either use more memory or reduce the effective sample rate. This is why a scope with an impressive maximum sample rate can provide far fewer samples per second during a long capture.
Memory matters for tasks such as:
- observing a power rail through a complete start-up sequence;
- capturing a long serial transaction while retaining individual edges;
- finding an intermittent glitch before a reset;
- correlating an analogue event with firmware or bus activity;
- examining switching behaviour over a changing load cycle.
Ask what sample rate remains available at the record length and time span you will actually use.
Memory also creates a navigation problem
A deep record is useful only if the instrument can search, zoom and measure it efficiently.
Look for hardware or software support for event search, serial decoding, mask testing, histograms and segmented memory. Segmented acquisition can retain short events separated by long inactive periods without filling the memory with uneventful samples.
Processing time matters too. A very deep acquisition that freezes the interface for several seconds can reduce the chance of catching an intermittent problem.
Waveform update rate affects rare-event detection
The sample rate describes points within an acquisition. Waveform update rate describes how many acquisitions can be processed and displayed over time.
A scope may sample a single record very quickly but spend much of the following second processing, transferring and drawing it. During that dead time, a rare glitch can pass unseen.
Manufacturers specify update rate in different modes and under particular conditions. Treat the maximum figure as a comparison input, then test the intended trigger, memory and analysis configuration.
The probe is part of the instrument
Connecting a probe changes the circuit. Input capacitance can slow an edge, alter an oscillator or increase loading on a high-impedance node. A long ground lead adds inductance and can create ringing that does not exist at the measurement point.
The complete measurement system includes the scope input, probe, accessories and connection method. Check:
- probe bandwidth and rise time;
- input capacitance and resistance;
- permitted common-mode and differential voltage;
- ground-lead inductance;
- attenuation accuracy;
- deskew between channels;
- whether the probe's dynamic range covers the signal.
For fast power measurements, a short ground spring, coaxial connection or suitable differential probe may change the result more than another bandwidth tier on the scope.
Vertical resolution and noise
An eight-bit converter has 256 nominal levels across the selected vertical range. If a small ripple sits on a large DC voltage, only a small portion of those levels may describe the ripple.
High-resolution acquisition, averaging and lower-noise front ends can improve useful detail, but their benefits depend on bandwidth and acquisition mode. Digital averaging cannot recover a one-off event and may hide variation.
Set the vertical scale so the signal uses the available display and converter range without clipping. Use an offset function where appropriate rather than placing a small AC feature inside an unnecessarily large range.
Effective number of bits and noise at the intended bandwidth are more informative than the nominal converter resolution alone.
Triggering determines what is captured
A stable display is not the same as a relevant capture. Edge triggering is sufficient for many tasks, but runt, pulse-width, timeout, logic, serial and zone triggers can isolate more specific failures.
Define the event before adjusting the instrument. For an intermittent reset, the useful trigger may be a power-rail dropout, a missing clock pulse or an invalid bus transaction. Capturing all three on correlated channels can turn a vague symptom into a sequence.
A practical buying and setup checklist
Before choosing a scope, write down:
- the fastest expected rise time;
- the longest event that must be captured;
- the number of simultaneous analogue and digital channels;
- the smallest voltage detail that matters;
- common-mode and differential-voltage requirements;
- serial buses or power measurements to decode;
- the rarest event that must be found;
- required probes and connection accessories.
Then calculate the approximate bandwidth, sample rate and memory combination. Verify those specifications with all required channels enabled.
When taking the measurement, record the probe, connection, bandwidth limit, sample rate, record length, acquisition mode and vertical scale. A screenshot without its acquisition conditions is weak engineering evidence.
The most useful oscilloscope is not necessarily the one with the largest number on its front panel. It is the instrument that preserves the detail, duration and probability of the event you need to understand.
Technical sources
- Tektronix, Evaluating oscilloscope bandwidth, sample rate and key specifications.
- Tektronix, Oscilloscope selection guide.
- Tektronix, sample-rate guidance and the relationship between record length, time and sample rate.
