diff --git a/src/iir.rs b/src/iir.rs
index 38edb4c..3b15885 100644
--- a/src/iir.rs
+++ b/src/iir.rs
@@ -1,23 +1,62 @@
+use core::f32;
+
 pub type IIRState = [f32; 5];
 
+#[derive(Default,Copy,Clone,Debug)]
 pub struct IIR {
-    pub y0: f32,
+    pub x_offset: f32,
+    pub y_offset: f32,
     pub ba: IIRState,
     pub scale: f32,
 }
 
 impl IIR {
+    pub fn pi(&mut self, kp: f32, ki: f32, g: f32) -> Result<(), &str> {
+        if ki < 0. {
+            return Err("ki < 0");
+        }
+        let (a1, b1, b0) = match () {
+            _ if ki < 0. => return Err("ki < 0"),
+            _ if ki < f32::EPSILON => (0., 0., kp),
+            _ => {
+                let (c, a1) = match () {
+                    _ if g < 0. => return Err("g < 0"),
+                    _ if g < f32::EPSILON => (1., 1.),
+                    _ => {
+                        let c = 1./(1. + ki/g);
+                        (c, 2.*c - 1.)
+                    }
+                };
+                let b0 = kp + ki*c;
+                let b1 = b0 - 2.*kp*c;
+                if b0 + b1 > -f32::EPSILON && b0 + b1 < f32::EPSILON {
+                    return Err("low integrator gain and/or gain limit");
+                }
+                (a1, b1, b0)
+            }
+        };
+        if b0 > 1. || b0 < -1. || b1 > 1. || b1 < -1. {
+            return Err("high gains");
+        }
+        self.ba[0] = b0;
+        self.ba[1] = b1;
+        self.ba[2] = 0.;
+        self.ba[3] = a1;
+        self.ba[4] = 0.;
+        Ok(())
+    }
+
     pub fn update(&self, xy: &mut IIRState, x0: f32) -> f32 {
         xy.rotate_right(1);
-        xy[0] = x0;
-        let y0 = macc(self.y0, xy, &self.ba, self.scale);
+        xy[0] = x0 - self.x_offset;
+        let y0 = macc(self.y_offset, xy, &self.ba)
+            .min(self.scale).max(-self.scale);
         xy[xy.len()/2] = y0;
         y0
     }
 }
 
-fn macc(y0: f32, x: &[f32], a: &[f32], scale: f32) -> f32 {
+fn macc(y0: f32, x: &[f32], a: &[f32]) -> f32 {
     y0 + x.iter().zip(a.iter())
         .map(|(&i, &j)| i * j).sum::<f32>()
-        .min(scale).max(-scale)
 }
diff --git a/src/main.rs b/src/main.rs
index 93e4419..d191b22 100644
--- a/src/main.rs
+++ b/src/main.rs
@@ -426,6 +426,8 @@ fn main() -> ! {
     rcc.apb1lenr.modify(|_, w| w.tim2en().set_bit());
     tim2_setup(&dp.TIM2);
 
+    unsafe { IIR_CH[0].pi(-0.01, 10.*2e-6/2., 0.).expect("bad coefficients"); }
+
     cortex_m::interrupt::free(|cs| {
         cp.NVIC.enable(stm32::Interrupt::SPI1);
         cp.NVIC.enable(stm32::Interrupt::TIM2);  // FIXME
@@ -434,22 +436,24 @@ fn main() -> ! {
     });
 
     loop {
-        cortex_m::asm::wfi();
+        for _ in 0..1000000 { cortex_m::asm::wfi(); }
+        let (x0, y0) = unsafe { (IIR_STATE[0][0], IIR_STATE[0][2]) };
+        info!("x0={} y0={}", x0, y0);
     }
 }
 
 #[interrupt]
 fn TIM2() {  // FIXME
     let dp = unsafe { Peripherals::steal() };
-    dp.TIM2.sr.modify(|_, w| w.uif().clear_bit());
+    dp.TIM2.sr.write(|w| w.uif().clear_bit());  // rc_w0
     dp.SPI1.cr1.write(|w| unsafe { w.bits(0x201) });
 }
 
 const SCALE: f32 = ((1 << 15) - 1) as f32;
 static mut IIR_STATE: [IIRState; 2] = [[0.; 5]; 2];
 static mut IIR_CH: [IIR; 2] = [
-    IIR{ y0: 0., ba: [1., 0., 0., 0., 0.], scale: SCALE },
-    IIR{ y0: 0., ba: [1., 0., 0., 0., 0.], scale: SCALE }];
+    IIR{ y_offset: 0., x_offset: 0., ba: [0., 0., 0., 0., 0.], scale: SCALE },
+    IIR{ y_offset: 0., x_offset: 0., ba: [0., 0., 0., 0., 0.], scale: SCALE }];
 
 #[interrupt]
 fn SPI1() {
@@ -458,8 +462,6 @@ fn SPI1() {
     cortex_m::interrupt::free(|cs| {
         let spi1p = SPI1P.borrow(cs).borrow();
         let spi1 = spi1p.as_ref().unwrap();
-        let spi2p = SPI2P.borrow(cs).borrow();
-        let spi2 = spi2p.as_ref().unwrap();
         let sr = spi1.sr.read();
         if sr.eot().bit_is_set() {
            spi1.ifcr.write(|w| w.eotc().set_bit());
@@ -468,12 +470,15 @@ fn SPI1() {
             // needs to be a half word read
             let rxdr1 = &spi1.rxdr as *const _ as *const u16;
             let a = unsafe { ptr::read_volatile(rxdr1) };
+            let x0 = a as i16 as f32;
+            let y0 = unsafe { IIR_CH[0].update(&mut IIR_STATE[0], x0) };
+            let d = y0 as i16 as u16 ^ 0x8000;
+
+            let spi2p = SPI2P.borrow(cs).borrow();
+            let spi2 = spi2p.as_ref().unwrap();
             // needs to be a half word write
             let txdr2 = &spi2.txdr as *const _ as *mut u16;
-            let x0 = a as f32;
-            let y0 = unsafe { IIR_CH[0].update(&mut IIR_STATE[0], x0) };
-            unsafe { ptr::write_volatile(txdr2, y0 as i16 as u16 ^ 0x8000) };
-            info!("adc={:.1} dac={:.1}", x0, y0);
+            unsafe { ptr::write_volatile(txdr2, d) };
         }
     });
     #[cfg(feature = "bkpt")]