Firmware 9.7.5 to 9.7.8: Four Fixes, Four Reports
Share
Four updates have gone out since 9.7.4. Every one of them started with a message from a user.
Some CC messages never arrived when you changed bank (9.7.5)
A 4ES2K owner on macOS told us that with BROADCAST ON BANK CHANGE enabled and four encoders on CC20 to CC23, changing bank delivered only CC20 and CC23. CC21 and CC22 never arrived, even though their stored values were intact and all four went out normally when he saved from the editor.
We measured the same unit on the same firmware, 30 rounds on each operating system. Windows lost none of 120 messages. macOS lost 20 of 120. The device was sending everything and the receiving side was dropping it. A single bank change can emit 19 messages back to back inside 0.1 ms, and CoreMIDI does not always keep up with a group that tight.
Bank changes now leave a very short gap between the messages they emit. macOS then lost 0 of 360. Nothing about the behaviour on Windows changes. About one bank change in a hundred still loses the whole group at once, which is a different fault with a different cause, and we are still looking at it.
A knob that kept sending the same value (9.7.6)
Three things in the potentiometer and fader path were fixed together.
-
Repeated values. Moving a knob slowly could send the same CC value two or three times in a row. Field logs showed runs like
cc 103, 103, 103. The decision to send was made on a finer internal reading than the value that actually goes out, so a few counts of noise opened the gate while the outgoing value stayed where it was, about half the time. After the fix we measured 3057 CC messages across a full sweep with zero consecutive duplicates. - Full scale on recent hardware. On boards built after our 4008 revision, 14-BIT CC, NRPN, Pitch Bend and Relative mode could not reach their maximum. A knob set to 14-BIT CC stopped near 15402 instead of 16383. All four modes now travel the whole range.
- Filter default raised from 4 to 9. Level 4 is fine for ordinary 7-bit CC but not for 14-BIT CC or NRPN, where one step of analog noise becomes a jump of 16. With hands off the controller for 10 seconds at level 4, the low byte wandered across a 32 count spread and 40 messages went out on their own. At level 9, none did. A full sweep at level 9 still reaches 0 to 16383, so the resolution is intact.
That new default applies to new units and to a factory reset only. If you already own a device, your setting is kept when you update. If you use 14-BIT CC or NRPN and see values drifting while you are not touching anything, raise the filter level yourself in the Web Configuration Tool under Global Variables.
Recalling a scene sent two values on one CC (9.7.7)
Another 4ES2K owner reported that recalling a scene sent the encoder's stored value and then the Virtual Fader value on the same CC. The receiving parameter passed through the stale number on its way, and his automation lane recorded both. The encoder's own position never moved either, so an encoder parked at 6 against a Virtual Fader of 64 would emit 7 on the next detent and the parameter fell off a cliff.
When a Virtual Fader sends a value now, any knob or encoder sharing that CC takes the value as its own position. The duplicate is gone as well. Measured on an 8EN with the encoder stored at 100 and the bank's Virtual Fader at 55, entering the bank used to emit 100, 55 and now emits 55 on its own.
Knobs felt slower than before (9.7.8)
An 8P2K owner wrote in after 9.7.6: "it works with 9, but it's slower. With 4, it's faster." He was right. The 9.7.6 measurement had only ever looked at a knob set to 14-BIT CC.
The filter settles after a number of samples rather than after a fixed time. Each knob is read once per pass of the main loop, so the more knobs a product has, the longer the same filter level takes to settle. At level 9 that is roughly 73 ms on four knobs, 146 ms on eight and 437 ms on twenty-four.
So the filter is now capped for knobs sending ordinary 7-bit CC, and the cap is divided by how many knobs the product has. Four or fewer keep 4, eight get 2, sixteen or more get 1. Response time stays roughly the same whichever product you own. Pitch Bend, 14-BIT CC and NRPN keep whatever level you set, for the reason 9.7.6 gives above.
These cap values are a starting point rather than a tuned result. If your knobs feel sluggish or noisy after updating, tell us and we will adjust them.
How to update
- Download the latest ParksTool Uploader from the releases page. Firmware updates need the desktop app; config.parkstool.com edits settings but cannot flash firmware.
- Run it, connect your device over USB, and press Auto Update. It detects your device and installs the firmware for it.
Your settings carry over. None of 9.7.5 through 9.7.8 changes the way the device stores its configuration, so your assignments, banks and colours stay where you left them.
If something does not behave the way you expect, let us know. All four of these started that way.