diff --git a/src/gateware/cxp_downconn.py b/src/gateware/cxp_downconn.py
index 5226f36..962560e 100644
--- a/src/gateware/cxp_downconn.py
+++ b/src/gateware/cxp_downconn.py
@@ -10,6 +10,113 @@ from artiq.gateware.drtio.transceiver.gtx_7series_init import *
 from operator import add
 from functools import reduce
 
+class CXP_DownConn(Module, AutoCSR):
+    def __init__(self, refclk, pads, sys_clk_freq, debug_sma, pmod_pads):
+        self.rx_start_init = CSRStorage()
+        self.rx_restart = CSRStorage()
+
+        self.tx_start_init = CSRStorage()
+        self.tx_restart = CSRStorage()
+        self.txenable = CSRStorage()
+
+
+        self.txinit_phaligndone = CSRStatus()
+        self.rxinit_phaligndone = CSRStatus()
+        self.rx_ready = CSRStatus()
+
+        # # #
+
+        # TODO: QPLL (GTXE2_COMMON) here
+        # CPLL too slow for 12.5Gbps :(
+
+        # single & master tx_mode can lock with rx in loopback 
+        self.submodules.gtx = CXP_GTX(refclk, pads, sys_clk_freq, tx_mode="single", rx_mode="single")
+        
+        self.comb += [
+            self.txinit_phaligndone.status.eq(self.gtx.tx_init.Xxphaligndone),
+            # self.rxinit_phaligndone.status.eq(self.gtx.rx_init.Xxphaligndone),
+            self.rx_ready.status.eq(self.gtx.rx_ready),
+
+            self.gtx.txenable.eq(self.txenable.storage[0]),
+            self.gtx.tx_restart.eq(self.tx_restart.storage),
+            self.gtx.rx_restart.eq(self.rx_restart.storage),
+            self.gtx.tx_init.clk_path_ready.eq(self.tx_start_init.storage),
+            self.gtx.rx_init.clk_path_ready.eq(self.rx_start_init.storage),
+            # self.gtx.rx_alignment_en.eq(self.rx_data_alignment.storage),
+        ]
+
+        # DEBUG:loopback 
+        self.loopback_mode = CSRStorage(3)
+        self.comb += self.gtx.loopback_mode.eq(self.loopback_mode.storage) 
+
+        # DEBUG:SMA
+        self.specials += [
+            Instance("OBUF", i_I=self.gtx.rxoutclk, o_O=debug_sma.p_tx),
+            Instance("OBUF", i_I=self.gtx.cd_cxp_gtx_tx.clk, o_O=debug_sma.n_rx)
+        ]
+
+        # DEBUG: datain
+        counter_max = 2
+        counter = Signal(max=counter_max)
+
+        self.data_0 = CSRStorage(8)
+        self.data_1 = CSRStorage(8)
+        self.data_2 = CSRStorage(8)
+        self.data_3 = CSRStorage(8)
+        self.control_bit_0 = CSRStorage()
+        self.control_bit_1 = CSRStorage()
+        self.control_bit_2 = CSRStorage()
+        self.control_bit_3 = CSRStorage()
+        self.encoded_0 = CSRStatus(10)
+        self.encoded_1 = CSRStatus(10)
+
+        self.rxdata_0 = CSRStatus(10)
+        self.rxdata_1 = CSRStatus(10)
+        self.decoded_data_0 = CSRStatus(8)
+        self.decoded_data_1 = CSRStatus(8)
+        self.decoded_k_0 = CSRStatus()
+        self.decoded_k_1 = CSRStatus()
+
+        
+        self.sync.cxp_gtx_tx += [
+            If(counter == 0,
+                self.gtx.encoder.d[0].eq(self.data_0.storage),
+                self.gtx.encoder.k[0].eq(self.control_bit_0.storage),
+                self.gtx.encoder.d[1].eq(self.data_1.storage),
+                self.gtx.encoder.k[1].eq(self.control_bit_1.storage),
+                counter.eq(counter+1),
+            ).Elif(counter == 1,
+                self.gtx.encoder.d[0].eq(self.data_2.storage),
+                self.gtx.encoder.k[0].eq(self.control_bit_2.storage),
+                self.gtx.encoder.d[1].eq(self.data_3.storage),
+                self.gtx.encoder.k[1].eq(self.control_bit_3.storage),
+                counter.eq(0),
+            ),
+            self.encoded_0.status.eq(self.gtx.encoder.output[0]),
+            self.encoded_1.status.eq(self.gtx.encoder.output[1]),
+        ]
+
+        self.sync.cxp_gtx_rx += [
+            self.rxdata_0.status.eq(self.gtx.decoders[0].input),
+            self.decoded_data_0.status.eq(self.gtx.decoders[0].d),
+            self.decoded_k_0.status.eq(self.gtx.decoders[0].k),
+
+            self.rxdata_1.status.eq(self.gtx.decoders[1].input),
+            self.decoded_data_1.status.eq(self.gtx.decoders[1].d),
+            self.decoded_k_1.status.eq(self.gtx.decoders[1].k),
+        ]
+
+
+        # The TX phase alignment will fail with a wrong TXUSRCLK frequency
+
+
+
+        # NOTE: No need to connect cxp_gtx_tx, we don't use tx anyway (just for loopback)
+
+        # TODO: Connect slave cxp_gtx_rx clock tgt
+        # checkout channel interfaces & drtio_gtx
+        # checkout GTPTXPhaseAlignement for inspiration
+
 # Changes the phase of the transceiver RX clock to align the comma to
 # the LSBs of RXDATA, fixing the latency.
 #
@@ -117,7 +224,7 @@ class CXP_BruteforceClockAligner(Module):
 
 
 
-class CXP_DownConn(Module):
+class CXP_GTX(Module):
     # Settings:
     # * GTX reference clock @ 125MHz
     # * GTX data width = 20