Comparative PID Analysis: Self Tuned profile vs. Jaiwen's 700-Class configs

This side-by-side comparison reveals some fascinating insights into why copy-pasting a 700-class tune directly onto a 550 can yield mixed results, but it also highlights the master-class control theory logic hidden in Jaiwen's choices.

The most striking initial difference is how Jaiwen treats Roll and Pitch asymmetrically compared to your symmetrical setup:

• Roll Proportional (Drop from 80 to 51): Jaiwen drastically lowers Roll P. On a 700, high roll inertia requires less aggressive proportional muscle to stop rotation. On your 550, dropping to 51 might feel a bit mushy unless compensated elsewhere.
• Pitch Proportional (Jump from 70 to 115): Conversely, Pitch P is bumped up significantly. This compensates for elevator pitching inertia and leverages higher bandwidth to snap stops.
• Bandwidth Shift: Look at the Roll Bandwidth (50 to 85) and Pitch Bandwidth (50 to 65). Jaiwen opens up the filter bandwidth significantly, allowing the controller to react much faster to high-frequency changes, which works hand-in-hand with his higher loop speeds.

The tail parameters show the most dramatic philosophical differences:

• Yaw Proportional (160 down to 125) & Integral (160 up to 350): You rely heavily on a high proportional gain (160) for immediate snap. Jaiwen drops P to 125 but more than doubles the I-term to an aggressive 350. This means his tune relies on the integral memory to "lock" the heading rather than brute-force proportional resistance.
• Stop Gains (CW 150/CCW 100 vs. CW 100/CCW 54): Your stop gains are much higher. Jaiwen's stops are softened significantly on paper, which prevents the tail from "bouncing" or barking on aggressive reversals, trusting the high I-term to finish the stop smoothly.

• Collective to Pitch Compensation (Off vs. On, Gain 10): Jaiwen runs active collective-to-pitch mixing to eliminate ballooning during heavy collective pumps, whereas yours was turned off.
• Cyclic Cross-Coupling (Gain 50 down to 5, Ratio 0 up to 10): You use a heavy cross-coupling filter (50), whereas Jaiwen strips it down to 5. This implies his physical swashplate setup or geometric decoupling is mechanically cleaner, or he prefers minimal software cross-coupling interference during rolls and loops.

Dropping Jaiwen's raw 700 PIDs straight into your 550 explains why certain elements felt unusual. However, you can cherry-pick his structural genius:

1. Keep Your Proven Cyclic P's (for now): Dropping Roll P down to 51 on a nimble 550 might make it feel loose.
2. Adopt His Bandwidth & Filtering Philosophy: His higher bandwidth settings (85/65/220) can safely sharpen your 550's response if your mechanical setup is clean.
3. Test His Tail Philosophy: Moving toward a higher I-term (like 250-300) with more moderate stop gains is a worthy experiment to eliminate tail bounce on hard stops.

That is a very smart boundary to draw. Leaving his raw proportional gains behind while focusing on his underlying control architecture-the structural "glue" like I-terms, error limits, cutoffs, and precomps-is where you will extract the true gold of a control theorist's design without risking an unflyable or over-twitchy machine.

Since your gyro filters are already locked in and performing great, and you want to bypass a blind copy-paste of his proportional loops, let's break down his non-PID structural parameters and devise a tactical integration plan.

• What Jaiwen did: Dropped Yaw Proportional to 125, massively pumped Yaw Integral to 350, and softened stop gains (CW 100 / CCW 54 vs. your 150/100).
• Why it works: Instead of relying on a stiff, twitchy proportional reaction to catch a tail stop (which can cause bounce or bark), his setup relies on the integration memory to hold position aggressively while feathering the stop dynamically.
• Integration Tactic: Adopt this now. On a 550, a high I-term paired with cleaner stops prevents tail oscillation during rapid reversals. Try porting his Tail I-term and stop gains directly; it scales safely across sizes because it governs persistence rather than raw velocity.

• What Jaiwen did: Tightened Error Limits across the board (Roll/Pitch down to 40, Yaw down to 45) and slightly shifted cutoff points (raising Roll cutoff to 15, Pitch to 12).
• Why it works: Error limits restrict how much cumulative error the integral term can wind up storing during extreme stick movements, preventing "integral windup" overshoot.
• Integration Tactic: Adopt cautiously. Tightening error limits to 40-45 on a 550 is a great way to keep the helicopter feeling crisp and prevent it from feeling "mushy" after a massive maneuver.

• What Jaiwen did: Bumped HSI limits up to 100 for Roll and Pitch while slightly reducing HSI Offset Gains to 45.
• Why it works: HSI ensures your cyclic response remains identical whether your headspeed is sagging under heavy load or running at full RPM. Letting the limit reach 100 gives the system full headroom to compensate for headspeed drops during hard collective pitches.
• Integration Tactic: Adopt this. This parameter is largely independent of helicopter size because it calculates relative percentage adjustments to headspeed changes.

• What Jaiwen did: Turned Collective to Pitch Compensation ON (Gain 10) and radically stripped down Cyclic Cross-Coupling Gain from 50 down to 5, while increasing the Cross-Coupling Ratio to 10.
• Why it works: Turning on collective-to-pitch mixing fixes the natural tendency of the heli to balloon during heavy collective climbs. Meanwhile, lowering software cross-coupling to 5 indicates his swashplate/head geometry has minimal mechanical coupling, or he prefers the raw mechanical reality over heavy software filtering.
• Integration Tactic: Test the Collective-to-Pitch Compensation immediately. If your 550 pitches slightly forward or backward during a heavy punch-out, turning this on with a modest gain of 10 will lock it flat. For cross-coupling, step yours down gradually (e.g., from 50 to 25, then down to 10) to see if your 550's geometry handles it cleanly without introducing unwanted roll during pure elevator inputs.

Instead of copying Profile #4 blindly, you can manually build a "Hybrid Profile" in one of your open slots:

1. Keep your preferred cyclic P-gains (or a mild blend).
2. Import his Tail parameters (High I-term, soft stops, yaw feedforward).
3. Import his Error Limits and Cutoffs to tighten the controller's boundaries.
4. Enable Collective-to-Pitch Compensation to stop any vertical ballooning.

How do you feel about moving these structural boundary layers and precomps into an active test bank first, leaving the raw proportional muscle closer to your organic baseline?

When you execute a collective pump, rising and falling in altitude is entirely intentional-you are commanding vertical movement. Ballooning, however, is an unwanted, non-linear pitching error that happens *during* that vertical move.

• What it looks like: You punch collective to climb rapidly, but instead of shooting straight up on a clean vertical vector, the helicopter's nose undesirably pitches up or down on its own due to aerodynamic downwash hitting the horizontal stabilizer or main rotor disc loading changes.
• The distinction: Intentional altitude change tracks precisely with your stick input. Ballooning is the surprise pitch deviation that forces you to mash forward or back on the cyclic just to keep the fuselage vertical during a pure vertical punch.

Cross-coupling occurs when an input meant for one axis accidentally causes movement on a completely different axis because of mechanical geometry.

• What it looks like: Imagine you are doing a pure, axial roll to the left. If you notice the nose of the helicopter stubbornly wanting to hook or pitch towards the belly or canopy instead of spinning cleanly on a flat center point, that is cross-coupling.
• The symptom: A roll or a loop that feels "spiraly" or uncoordinated rather than mechanical and locked to a single plane. Heavy software cross-coupling filters try to mask this electronically, whereas cleaner geometry (or lower interference settings) lets the physical airframe track true.

Testing every single parameter one-by-one would indeed take forever and drive you crazy. Instead of micro-testing, we can bundle Jaiwen's non-PID architecture into three logical packages based on what they actually achieve. This cuts your flight testing down to just three distinct test flights.

Recommendation: Skip testing individual cutoffs or single parameters. Apply Package A and the Governor Precomps for your next flight. They are safe, size-independent structural improvements that directly target that ultra-crisp, locked-in feel without messing with your proportional muscle.

Your observations are spot on, and they completely align with why you didn't feel a massive transformation between profiles:

You hit the nail on the head regarding scale. On a 700mm machine, the massive rotational inertia means the blades take longer to spin up or slow down during rapid direction changes, making I-term windup a major factor that requires heavy relaxation. On a nimble 550mm helicopter, the natural recovery is much faster, so a subtle shift in I-term relax parameters won't radically alter the everyday feel during standard flips and rolls.

As you noted, small incremental shifts like shifting Error Limits or HSI Offset limits/gains by just 5 points are subtle tuning layers. They act as fine-tuning "guardrails" rather than primary behavioral switches. If your baseline was already flying well, a 5-point shift won't pop out as a night-and-day difference during moderate-to-hard stick jabs.

Since Package A's architectural limits didn't change the game, we know your baseline stability is already strong. To find that next level of "locked-in" performance, the real performance-drivers in Jaiwen's setup are likely hiding in two other places we haven't touched yet:

• The Tail Architecture (High I-term & Stop Gains): Tail parameters yield a very distinct behavioral change on hard stops and reverses.
• Governor Precomps (Collective Precomp bumped to 100): This is massive for maintaining strict headspeed consistency during high-load maneuvers on smaller machines.

Would you like to shift your focus to testing his tail configuration or look at the Governor precomp side next?

I completely agree with your assessment. Characterizing limits, cutoffs, and bandwidth parameters as architectural "configuration guardrails" rather than aggressive behavioral levers is an accurate way to view them.

Parameters like Error Limits, HSI Offset limits/gains, and minor cutoff shifts operate in the background:

• They are boundaries, not commands: Unlike P-gains, Feed-Forward, or Boost-which actively inject force or speed into the control loop every time you move a stick-limits and cutoffs only wake up when the system approaches an extreme edge case (like saturation or high-frequency noise).
• Subtle deltas mean subtle life: A +/- 5 change on a limit means the flight controller's math shifts slightly right before it clips an error accumulator. During normal, precise flight, you may never even hit those thresholds, which is why your routine of rolls, flips, and jabs felt virtually identical.

If you want parameters that actually change the DNA of how the helicopter feels in flight, you have to look at the parts of the tune that actively manipulate control delivery:

• Precomps and Mixing: (Like turning Collective-to-Pitch compensation on or maxing out Governor Collective precomp to 100), which actively push feed-forward compensation into the mechanics before physics even forces a sag or balloon.
• Tail Architecture: (Like shifting Yaw Integral to 350), which radically alters how persistence and holding power behave on hard stops.

By treating those boundary parameters as set-and-forget safety configurations, you've essentially locked down a very clean, stable operational envelope. Now you can focus your testing energy on the parameters that actually bite!