Remote Heads, SDR and CW: Where the Delay Really Matters
Remote Heads, SDR and CW: Where the Delay Really Matters
A radio without a front panel does not have to feel distant. A well-designed network station can tune naturally, sound clean, and work very well on CW. The useful question is which part of the operation is delayed—and whether that delay reaches your ears, your transmitted signal, or just the screen.
The short answer: yes, it can feel like a normal radio
With a responsive application, sensible audio buffers and a stable connection, the extra delay of remote operation can be small enough that you do not notice it during ordinary listening and conversation. You turn the dial, find the station and get on with the contact. There is no general rule that a networked radio must feel sluggish.
But “I do not notice it while listening” is different from “there is no delay.” Listen to the same station through two different receivers and a modest timing difference can become obvious. Listen to your own Morse coming back from a distant transmitter and a delay that was harmless during reception can become distracting.
For comfortable remote CW, keep the immediate hand-to-ear feedback local. Let the remote system handle transmission timing properly. A good Internet connection helps, but local sidetone solves a different problem from a faster connection.
The head has moved, but the architectures are different
A control head is the part with the display, tuning controls and operator connections. An SDR, or software-defined radio, performs significant radio functions through digital processing. Neither label says exactly where that processing happens. It might be inside the radio, in a computer, in the head, or divided among them.
| Example | Where the operator works | What to include in the delay budget |
|---|---|---|
| SunSDR / ExpertSDR3 | A computer provides the software operating panel; remote operation uses the appropriate station-side and client software. | The computer, its audio path, the network and the selected remote configuration. See ExpertSDR3. |
| ANAN / Thetis | In a PC/Thetis arrangement, the computer contributes DSP, display and control. ANAN models and onboard processing arrangements vary. | DSP and audio buffers, filter choices, hardware/firmware and the actual keying route. See the manufacturer-hosted Thetis manual. |
| FlexRadio / SmartSDR / Maestro | Software clients or a Maestro console operate a networked radio; SmartLink provides Internet access. | Radio processing, client audio, network conditions and the chosen CW interface. See SmartLink setup. |
| Icom IC-7760 | The radio consists of a separate physical head and RF deck. The current system supports LAN and Internet operation. | Processing in the head and deck, the connection and adaptive buffering. The supplied head is part of this two-piece system; an additional head is a separate option. See Icom’s current description. |
A box full of knobs is not proof of zero latency. A headless box is not proof of high latency. Even a conventional desktop receiver has filters and processing with a finite response time. The actual signal path matters more than the shape of the enclosure.
Separate the delays before comparing the numbers
Control response is the time between turning a dial and the radio applying the change. The display may update before the radio acknowledges it. The newly tuned audio then has to travel back through the receive path.
Receive-audio latency runs from a signal arriving at the radio’s antenna input to the corresponding sound reaching your headphones. Sampling, filtering, demodulation, audio packaging, the network and playback can all contribute.
Transmit latency runs from your microphone or keying action to the corresponding RF output. Sidetone latency runs from your keying action to the tone you hear yourself. Those are separate measurements, even when both begin with the same paddle.
Across the desk, across the house, or across the Internet?
On a direct cable or an uncongested wired local network, the network’s contribution can be small compared with the radio’s filters and audio buffers. Moving the head into another room need not add a noticeable operating penalty.
Wi-Fi adds a shared radio link. It can work very well, but interference, competing traffic, retransmissions and weak coverage can make arrival times less predictable. A wired connection is a useful reference when diagnosing delays or dropouts.
Across the Internet, the route and both access connections join the calculation. A nearby station on a stable route can feel quite natural. A distant station, an overloaded uplink or a relay route may need more buffering. Geographic distance imposes a lower limit, but the shortest route on a map is not necessarily the route the packets take.
For ordinary reception, a steady delay simply means you hear the complete signal slightly later. That does not automatically damage the audio or make Morse harder to copy. For turn-taking, pileups and hearing an interruption during your own transmission, the accumulated delays can affect when you react.
What “a decent Internet connection” actually means
The number advertised on a broadband package measures capacity, not responsiveness. A lightly loaded, stable connection can serve a station better than a much faster connection with long queues.
- Enough capacity in both directions. The station must send receive audio to you; your location must send transmit audio or keying back. A wide spectrum or raw sample stream may require much more capacity than a demodulated audio stream.
- Low, stable delay. Check the route you actually use, including any VPN or relay, at the times you operate.
- Little jitter and loss. Jitter means variation in packet arrival timing. Audio cannot pause whenever one packet is late, so a receiver often holds a small reserve of samples.
- Short queues under load. A backup or video upload can fill the upstream queue even when a speed test looks excellent. Sensible traffic management can help keep interactive packets moving.
- A sound local audio path. Oversized sound-device buffers or a Bluetooth speaker can dominate the delay even when the network is behaving well.
The reserve used to absorb jitter is a tradeoff: a larger buffer tolerates more variation but makes the sound older. RTP’s timing and jitter definitions help separate these quantities; they do not promise that every packet will arrive on time.
A ping measures a round trip for its own packets. It does not include the complete radio-to-headphone path. Dividing it by two is only a rough one-way estimate when the two directions behave similarly. Also test while the connection is busy: an excellent idle ping can hide a large delay under load.
CW: delayed sound is not the same as damaged Morse
A clean Morse signal shifted by a constant 60 milliseconds still contains the same dots, dashes and spaces. An operator listening to that signal can usually copy it normally. The delay becomes more troublesome when the operator is trying to generate the timing while listening to that delayed sound.
With ordinary PARIS word-speed timing, a dot lasts 1,200 divided by the speed in words per minute, in milliseconds. The standard dot/dash/spacing relationships are specified in ITU-R M.1677; the PARIS convention is described in ARRL’s Morse-timing note.
| Speed | One dot | A 60 ms delay expressed in dots |
|---|---|---|
| 20 WPM | 60 ms | 1 dot |
| 30 WPM | 40 ms | 1.5 dots |
| 40 WPM | 30 ms | 2 dots |
Those ratios are a scale for understanding the problem, not an acceptable-delay standard. Different operators and tasks have different tolerances. They explain why a delay that barely attracts attention during listening may be very unwelcome in paddle feedback.
Generate the sidetone beside the operator
A local keyer can generate the tone as you operate the paddle. You hear what your hand is doing without waiting for information to reach the station and audio to return. The remote transmitter can reproduce the corresponding keying after the necessary transport and scheduling delay.
This is an established approach. The FlexRadio Maestro guide describes a local keyer and sidetone, together with buffering to accommodate network jitter. The buffer delays transmission while protecting its timing.
Local sidetone confirms the local keying action; it is not proof that RF left the remote antenna. Connection status, transmitter state and the remote system’s fault handling still matter.
Preserve the rhythm, not just the key-down message
A simple system that keys immediately whenever a network message arrives can turn changing network delays into changing dot lengths. A suitable remote-CW design instead uses a supported timing-aware protocol, a scheduling buffer, or a keyer at the station that generates Morse from higher-level commands.
Those are different solutions. Sending text to a remote keyer preserves its generated timing but does not offer the same interaction as forwarding a straight key or paddle. Timestamped events require an agreed timing reference and handling for late or missing events. The selected client and radio must support the same method.
The station also needs a defined response to a broken connection, including releasing key-down or PTT. A local tone continuing to play cannot substitute for that behavior.
Full break-in still has a timing budget
Full break-in, often called QSK, lets you hear between transmitted Morse elements. It depends on transmitter sequencing, receiver recovery, filtering and the audio path. Local sidetone does not remove any of those delays. Audio arriving between your local paddle elements may describe an earlier interval at the remote station.
Remote CW can be comfortable and effective without matching the feel of a very fast local QSK station. If hearing another operator between individual elements is central to how you operate, measure that complete behavior. For ordinary exchanges and semi break-in, a modest fixed delay is often much less important.
SmartCW: put the immediate feedback at your hand
SmartCW illustrates the operator end of this arrangement. Its local hardware keyer and sidetone let you use a paddle without hearing yourself through a distant receiver’s audio stream. The device connects over USB and offers interfaces including MIDI and WinKeyer support, alongside its local keying functions.
For remote operation, pair that local interface with client software that supports the chosen input mode and a suitable path to the transmitter. Keyed MIDI events, raw paddle events and buffered text are not interchangeable. A compatible application must interpret the selected mode correctly.
Mute any delayed returned transmit sidetone so that you hear one local tone. The useful benefit is immediate local feedback. SmartCW itself is not an Internet transport, and its sidetone does not cancel the remote receiver’s delay. The complete combination—device mode, client, network protocol and radio—determines how the transmitted CW is timed. Use the project documentation for the supported interfaces rather than assuming that any USB connection completes a remote station.
VFO-Knob: one dial, two receivers, two audio histories
The VFO-Knob Icom firmware provides a concrete example. A supported Icom transceiver can remain the transmit/receive radio while a separate KiwiSDR, Web-888 or UberSDR receiver follows its tuning. The Icom audio can be heard in one ear and the second receiver in the other, with balance control; the secondary SDR audio is muted during transmission.
That gives the operator another receiving location or antenna without giving up the familiar dial. However, following frequency, mode and passband does not align the audio in time. The Icom and the secondary SDR can have different filters, buffers, sample clocks and network routes. A remote receiving site also hears a different RF path.
If both receive the same station, a timing difference may produce a distracting double image. Selecting one stream or adjusting the balance may be preferable to listening equally to both. Do not describe this as phase-coherent diversity reception.
In a setup that does return remote-SDR audio during transmission, that delayed monitoring path should not be used as paddle sidetone. It includes the transmitter, the RF path, another receiver and its return audio transport. The VFO-Knob site covers the available radio and receiver modes.
OpenX6X00: browser operation and Bluetooth control
OpenX6X00 brings another arrangement to the Xiegu X6100 and X6200: a radio with its own physical controls can also be operated through its web interface. The project documents browser control with receive/transmit audio, network access and remote-access options.
A browser is an operator interface, not a fixed latency penalty. Its audio scheduling, the radio’s processing and the connection still determine the result. A browser on the same LAN and a browser reaching the station through an Internet relay are different timing cases, despite showing the same controls.
The project’s Bluetooth CAT connection is a separate control path. CAT means computer-aided transceiver control: commands for settings such as frequency and mode. Bluetooth control does not imply that all receive and transmit audio travels over Bluetooth, or that it has the timing of a Bluetooth headset.
This distinction also matters at the listening end. The VFO-Knob Bluetooth audio guide describes extra speaker buffering compared with the wired jack. A Bluetooth speaker can be convenient for casual listening; wired headphones are a better reference for CW latency tests. A quick tuning command and noticeably delayed speaker audio can coexist.
Putting numbers on a receive path
The following budgets are illustrative calculations, not measurements of the named products. They show how separate contributions can add up. All entries refer to one-way receive audio, from RF at the receiver to audio playback.
| Contribution | Wired LAN example | Stable regional Internet example | More buffered Internet example |
|---|---|---|---|
| Radio processing and filtering | 8 ms | 8 ms | 8 ms |
| Packet formation / sender buffering | 10 ms | 10 ms | 10 ms |
| One-way network transit | 1 ms | 20 ms | 50 ms |
| Receive jitter buffer | 10 ms | 20 ms | 80 ms |
| Playback path | 6 ms | 6 ms | 12 ms |
| Total | 35 ms | 64 ms | 160 ms |
The stable Internet example adds 29 ms to the LAN example. During ordinary listening, that difference may be operationally unobtrusive. It is not a prediction for your equipment or a claim that 64 ms is imperceptible in every task. Changing the filter or audio device can alter the comparison substantially.
For conversation, add the relevant transmit and receive delays when considering turn-taking. For sidetone returned from the station, include the outward keying path as well as the return audio. A one-way receive number cannot describe either complete loop.
ITU-T G.114 discusses conversational one-way delay and explicitly recognizes that highly interactive tasks can be affected at relatively low delays. Its telephony guidance is not a CW sidetone or QSK acceptance limit.
A deeper look: samples, filters and event timing
Buffers are time stored as samples
At a sample rate fs, a block of N samples spans:
At 48,000 samples per second, 480 samples span 10 ms and 960 span 20 ms. Depending on how blocks are filled, queued and scheduled, the latency of a particular sample can differ from a full block interval. Count the actual queues and avoid counting the same buffer twice.
Capacity is a different calculation. Uncompressed 48 kHz, 16-bit mono audio contains 768 kbit/s before protocol overhead. A raw 192 ksample/s stream containing two 24-bit I/Q components contains 9.216 Mbit/s before overhead. I/Q are the two numerical components used to represent the radio signal. These examples explain why audio-only and raw-sample remote systems can have very different network requirements; they are not the stream formats or minimum connection speeds of the products above.
Filtering has a time response, too
For a single-rate, linear-phase FIR filter with N taps at sample rate fs, the group delay is:
Thus 2,049 taps at 48 kHz give approximately 21.33 ms. This is a calculation for that filter model, not a complete receiver specification. Multirate filters need their delays referred to the correct rates; other filter types have different phase and delay behavior. See the linear-phase FIR group-delay relationship.
Sharper filter transitions can require longer processing histories. Some receivers let the operator trade filter sharpness against latency; FlexRadio documents that choice in its receive-filter guidance. That tradeoff exists before Internet transport is added. A very narrow filter may make a crowded-band signal easier to read while changing the timing of its audio.
Fixed delay cancels from an interval; jitter does not
Suppose a key-down event is generated at time t1 and key-up at t2. A simple arrival-driven transmitter sees delays d1 and d2. Its transmitted element lasts:
If both events are delayed equally, the element is merely shifted in time. If key-up arrives 15 ms later relative to key-down, a nominal 40 ms dot becomes 55 ms in this simplified example. A scheduling buffer can restore the intended interval when events arrive before their deadlines. It cannot make arbitrary losses or arbitrarily late packets disappear.
Similarly, delaying the faster of two audio streams can align them only if the difference is known and adequately stable. Independent sample clocks can drift. Matching two VFO readings is not a substitute for timing and clock alignment.
Measure the operation you care about
Start with wired headphones and a wired network where practical. Establish that reference, then add Wi-Fi, Internet access or Bluetooth one change at a time.
- For tuning: observe both the command response and when the newly tuned signal reaches the audio output. A smoothly moving number alone is insufficient.
- For receiving: apply a controlled, properly attenuated RF tone burst and compare its known timing with the audio output on the same recording or acquisition time base. A two-receiver recording can show relative delay, but does not establish either receiver’s absolute delay unless one path is calibrated.
- For CW: measure key action to local sidetone separately from key action to RF envelope. Use a suitable RF coupler/detector for the latter, not a direct connection from a transmitter output to an audio or SDR input.
- For break-in: check what you can hear between elements and after releasing the key, with the actual filter, transmit sequencing and remote-audio configuration.
- For network consistency: repeat under realistic upload/download load. Record typical delay, occasional long delays, packet loss and dropouts. A single best-case result is not the operating experience.
Record the radio and firmware, client, mode, filter width, audio device, buffer settings, connection type and both timing endpoints. Those details make a latency number useful to another operator.
Build the station around the right feedback path
For much ordinary HF operating, a modern remote head or software client on a good connection can feel entirely natural. A stable receive delay does not stop you copying a weak station. Moving the radio to a quieter location may bring a far larger practical benefit than the small extra delay costs.
CW deserves deliberate choices: immediate local sidetone, a remote keying method that preserves timing, and a receive path suited to the kind of break-in you want. Multiple receivers deserve separate timing expectations. Bluetooth control and Bluetooth audio deserve separate treatment.
Choose the architecture that fits the way you operate, then measure the complete path. A front panel is a user interface. Responsiveness comes from everything behind it.
Mini-FAQ
- Is every SDR slower than a conventional radio? No. Processing location, filters, buffers and audio routing determine latency. The SDR label or the presence of physical controls does not give an end-to-end number.
- Can remote operation feel as responsive as sitting at the radio? Yes, for many ordinary listening and conversational tasks on a well-configured connection. This is an operating experience, not a claim of zero delay or identical high-speed QSK behavior.
- Does low ping prove low audio latency? No. Ping measures its own round trip. Radio processing, packet formation, jitter buffers and audio playback add separate delays.
- Does SmartCW local sidetone remove Internet delay? It removes the network round trip from the local paddle-feedback path. The compatible client, remote keying and received-audio paths retain their own delays.
- Does a second SDR following my radio have synchronized audio? Not automatically. Frequency tracking does not align network buffers, sample clocks, receiver processing or RF propagation paths.
- Is Bluetooth control the same as Bluetooth audio? No. Sending CAT commands and transporting continuous audio are different functions. Audio buffering can add delay even when control feels responsive.
Found this useful? Support my writing and experiments: Buy me a coffee.